Evaluation of Real Integrals

How can we compute an integral like ∫−∞∞cos⁡xx2+1 dx\int_{-\infty}^\infty \frac{\cos x}{x^2+1}\,dx without antiderivatives? Real calculus offers no elementary antiderivative for such integrands. Yet the answer is the strikingly clean π/e\pi/e. The trick is to leave the real line entirely. View the integrand as the restriction of a meromorphic function on C\mathbb{C}. Close the path of integration with an arc. Let the Residue Theorem convert an intractable limit into a finite sum of residues. This is one of the most spectacular payoffs of complex analysis. It matters for working mathematicians, physicists, and engineers, because it turns hard analytic estimates into algebra. The big picture is a dictionary between real integrals and closed contour integrals in the plane. A real integral over (−∞,∞)(-\infty,\infty), over (0,2π)(0,2\pi), or over (0,∞)(0,\infty) with a branch cut, gets reinterpreted as one piece of a closed contour. The remaining pieces (large semicircles, small indentations around singularities, arcs avoiding branch cuts) are shown to vanish or to contribute computable amounts. This uses the ML Inequality and Estimates and its refinements (Jordan's Lemma, the small-arc lemma). What remains is 2πi2\pi i times a sum of residues, by the Residue Theorem. This is an algebraic object computed from Laurent series or simple pole formulas. This article organizes the subject around four standard contour families. These are: rational functions of cos⁡θ,sin⁡θ\cos\theta,\sin\theta over [0,2π][0,2\pi] (unit circle contours); rational and rational-times-trigonometric functions over R\mathbb{R} (semicircular contours, using Jordan's Lemma for oscillatory integrands); integrands with branch points (keyhole contours); and integrands with poles on the real axis (indented contours, via the small-arc lemma). Each family reduces a genuinely hard real computation to routine residue bookkeeping. We give the general machinery as a single unifying theorem. Then we work through the machinery on two asymmetric examples. These are chosen so no cancellation or symmetry trick can silently paper over an error. We also show, with an explicit failure, why the growth-and-decay hypotheses in Jordan's Lemma cannot be dropped.

Key takeaways
  • Real integrals that resist elementary antiderivatives often extend to meromorphic functions whose closed-contour integral is computable by residues.
  • The key engineering step is choosing a contour on which the "extra" pieces (arcs, indentations) provably vanish or contribute known amounts.
  • Jordan's Lemma upgrades the ML estimate for oscillatory factors eiaze^{iaz}, letting semicircular arcs vanish even when ∣f(z)∣|f(z)| only decays like 1/∣z∣1/|z|.
  • Keyhole contours handle branch points; indented contours handle poles that sit exactly on the integration path.
  • Every method reduces, in the end, to the Residue Theorem: ∮Cf=2πi∑Res⁡\oint_C f = 2\pi i \sum \operatorname{Res}.
  • Key idea: trade an unreachable real limit for a closed loop in C\mathbb{C} whose value is forced by the algebra of residues.

Where this lives today

Residue methods for real integrals are the standard toolkit in theoretical physics, where Green's functions, propagators, and dispersion relations (e.g. the Kramers–Kronig relations in optics) are literally Fourier-type integrals of rational or meromorphic functions, evaluated by closing contours in the upper or lower half-plane exactly as in Example 2. In signal processing and electrical engineering, computing inverse Laplace and Fourier transforms of rational transfer functions is routine residue calculus; stability analysis of linear systems depends on locating poles relative to a contour. In probability and statistics, characteristic functions of stable and Cauchy-type distributions are evaluated via contour integration, and saddle-point/steepest-descent methods (a refinement of these ideas) underlie asymptotic approximations in statistical mechanics and large-deviations theory. In analytic number theory, contour shifting and residue extraction (e.g. Perron's formula, the proof outline behind the Prime Number Theorem) generalizes the keyhole and semicircle ideas to integrals of the Riemann zeta function. In algorithms and computer algebra, symbolic integration packages recognize rational and rational-trigonometric integrands and dispatch to residue-based evaluation rather than searching for elementary antiderivatives.

Ask the tutor: How do I decide, for a specific integrand, whether to use plain ML, Jordan's Lemma, a keyhole contour, or an indentation — and can more than one be needed in the same problem?

DefinitionContour-integral evaluation of a real integral

Let I=∫abg(x) dxI = \int_a^b g(x)\,dx (possibly improper, a=−∞a=-\infty or b=∞b=\infty) be a real integral. A contour-integral evaluation of II consists of: 1. A meromorphic function ff on a domain Ω⊆C\Omega \subseteq \mathbb{C}, with f(x)=g(x)f(x) = g(x) for real x∈(a,b)x \in (a,b) (or gg recovered from ff via a fixed algebraic identity, e.g. cos⁡θ=12(z+z−1)\cos\theta = \tfrac12(z+z^{-1}) on ∣z∣=1|z|=1). 2. A family of closed, piecewise-C1C^1 contours CRC_R (or CR,εC_{R,\varepsilon}) bounding a region containing finitely many poles of ff, none lying on CRC_R itself, such that one arc or edge of CRC_R converges (as R→∞R\to\infty or ε→0\varepsilon \to 0) to the path of II. 3. Estimates (via the ML inequality, Jordan's Lemma, or the small-arc lemma) showing the remaining pieces of CRC_R tend to 00, to ±iπRes⁡\pm i\pi\operatorname{Res}, or to another explicit limit. The Residue Theorem then gives

∮CRf(z) dz=2πi∑zk inside CRRes⁡(f,zk),\oint_{C_R} f(z)\,dz = 2\pi i \sum_{z_k \text{ inside } C_R} \operatorname{Res}(f, z_k),

and letting R→∞R\to\infty (or ε→0\varepsilon\to 0) equates II, possibly combined with the vanishing/known pieces, to this residue sum.

Intuition

Think of the real line as a river you cannot ford directly — the integral has no closed form as an antiderivative. Complex analysis builds a bridge: extend the integrand to the whole plane (or a slit plane), and notice that a closed loop integral is rigid, forced to equal 2πi2\pi i times a sum of local residue data, no matter how wild the loop's shape, by the Residue Theorem. The strategic problem is purely geometric: design a loop that contains a piece equal to (or convergent to) your original integral, plus auxiliary pieces you can kill with estimates. For ∫−∞∞dx1+x2\int_{-\infty}^\infty \frac{dx}{1+x^2}, the natural loop is a large semicircle in the upper half-plane: the diameter is the real segment [−R,R][-R,R], and the arc ΓR\Gamma_R of radius RR closes the loop. Since f(z)=1/(1+z2)f(z) = 1/(1+z^2) decays like 1/R21/R^2 on ∣z∣=R|z|=R, the ML inequality kills the arc's contribution as R→∞R\to\infty, leaving exactly the residue at the single pole z=iz=i inside. When an oscillatory factor eiaxe^{iax} (a>0a>0) is present, decay like 1/R21/R^2 is not needed — Jordan's Lemma shows that 1/R1/R decay already suffices, because eiaze^{iaz} itself decays exponentially off the real axis in the upper half-plane. This is the mechanism behind Fourier-type integrals such as ∫cos⁡(ax)/(x2+1) dx\int \cos(ax)/(x^2+1)\,dx.

:::svg[Semicircular contour for ∫−∞∞dx/(1+x2)\int_{-\infty}^\infty dx/(1+x^2): diameter [−R,R][-R,R] plus arc ΓR\Gamma_R, enclosing the pole at z=iz=i] z = i (pole) -R R Γ_R (arc) [-R,R] (real axis)

TheoremResidue evaluation via semicircular contours, with Jordan's Lemma

Let ff be a function meromorphic on C\mathbb{C} except for finitely many poles, none on R\mathbb{R}, and let a≥0a \geq 0. Suppose: (i) ff has finitely many poles z1,…,znz_1,\dots,z_n in the open upper half-plane {Im⁡z>0}\{\operatorname{Im} z > 0\}; (ii) ff is continuous on R\mathbb{R}; (iii) If a=0a = 0: ∣f(z)∣≤MR|f(z)| \leq M_R on the semicircle ∣z∣=R|z|=R, Im⁡z≥0\operatorname{Im} z \geq 0, with MR→0M_R \to 0 as R→∞R\to \infty, and RMR→0R M_R \to 0. (iv) If a>0a > 0: f(z)→0f(z) \to 0 uniformly as ∣z∣→∞|z|\to\infty on Im⁡z≥0\operatorname{Im} z\geq 0 (i.e. max⁡∣z∣=R,Im⁡z≥0∣f(z)∣=:MR→0\max_{|z|=R,\operatorname{Im}z\ge0}|f(z)| =: M_R \to 0). Then, provided RR is chosen larger than ∣zk∣|z_k| for all kk and the improper integral on the left converges,

∫−∞∞f(x)eiax dx=2πi∑k=1nRes⁡(f(z)eiaz, zk).\int_{-\infty}^{\infty} f(x)e^{iax}\,dx = 2\pi i \sum_{k=1}^n \operatorname{Res}\big(f(z)e^{iaz},\,z_k\big).

In particular (taking a=0a=0 and using hypothesis (iii)),

∫−∞∞f(x) dx=2πi∑k=1nRes⁡(f,zk).\int_{-\infty}^\infty f(x)\,dx = 2\pi i \sum_{k=1}^n \operatorname{Res}(f,z_k).
LemmaJordan's Lemma

Let a>0a>0 and let ΓR\Gamma_R denote the semicircular arc z=Reiθz = Re^{i\theta}, θ∈[0,π]\theta \in [0,\pi]. If gg is continuous on ΓR\Gamma_R for all large RR and MR:=max⁡θ∈[0,π]∣g(Reiθ)∣→0M_R := \max_{\theta\in[0,\pi]} |g(Re^{i\theta})| \to 0 as R→∞R \to \infty, then

lim⁡R→∞∫ΓRg(z)eiaz dz=0.\lim_{R\to\infty} \int_{\Gamma_R} g(z) e^{iaz}\,dz = 0.
ProofProof of Jordan's Lemma

Technique: direct parametrization plus the elementary inequality sin⁡θ≥2θπ\sin\theta \geq \frac{2\theta}{\pi} on [0,π/2][0,\pi/2]. Parametrize z=Reiθz=Re^{i\theta}, dz=iReiθ dθdz = iRe^{i\theta}\,d\theta. Then

∣∫ΓRg(z)eiaz dz∣≤∫0π∣g(Reiθ)∣ ∣eiaReiθ∣ R dθ=∫0π∣g(Reiθ)∣ e−aRsin⁡θ R dθ,\left|\int_{\Gamma_R} g(z)e^{iaz}\,dz\right| \le \int_0^\pi |g(Re^{i\theta})|\,|e^{iaRe^{i\theta}}|\,R\,d\theta = \int_0^\pi |g(Re^{i\theta})|\, e^{-aR\sin\theta}\, R\,d\theta,

since eiaReiθ=eiaRcos⁡θe−aRsin⁡θe^{iaR e^{i\theta}} = e^{iaR\cos\theta}e^{-aR\sin\theta} has modulus e−aRsin⁡θe^{-aR\sin\theta}. Bounding ∣g(Reiθ)∣≤MR|g(Re^{i\theta})| \le M_R,

∣∫ΓRg(z)eiaz dz∣≤MRR∫0πe−aRsin⁡θ dθ=2MRR∫0π/2e−aRsin⁡θ dθ,\left|\int_{\Gamma_R} g(z)e^{iaz}\,dz\right| \leq M_R R \int_0^\pi e^{-aR\sin\theta}\,d\theta = 2 M_R R \int_0^{\pi/2} e^{-aR\sin\theta}\,d\theta,

using symmetry of sin⁡θ\sin\theta about θ=π/2\theta = \pi/2. On [0,π/2][0,\pi/2] the elementary inequality sin⁡θ≥2θπ\sin\theta \geq \dfrac{2\theta}{\pi} holds (concavity of sin⁡\sin on this interval together with equality at endpoints θ=0,π/2\theta=0,\pi/2 forces the chord 2θ/π2\theta/\pi to lie below the graph). Hence e−aRsin⁡θ≤e−2aRθ/πe^{-aR\sin\theta} \leq e^{-2aR\theta/\pi}, so

2MRR∫0π/2e−aRsin⁡θ dθ≤2MRR∫0π/2e−2aRθ/π dθ=2MRR⋅π2aR(1−e−aR)≤πMRa.2M_R R \int_0^{\pi/2} e^{-aR\sin\theta}\,d\theta \leq 2 M_R R \int_0^{\pi/2} e^{-2aR\theta/\pi}\,d\theta = 2M_R R \cdot \frac{\pi}{2aR}\Big(1 - e^{-aR}\Big) \leq \frac{\pi M_R}{a}.

Since MR→0M_R \to 0 and a>0a>0 is fixed, the right side tends to 00, proving the claim. ■\blacksquare

∎
ProofProof of the main theorem

Technique: closed semicircular contour + Residue Theorem + ML/Jordan estimate on the arc. Fix RR larger than max⁡k∣zk∣\max_k |z_k|, so that all poles z1,…,znz_1,\dots,z_n lie strictly inside the closed contour CRC_R consisting of the segment [−R,R][-R,R] on the real axis followed by the arc ΓR={Reiθ:0≤θ≤π}\Gamma_R = \{Re^{i\theta}: 0\le\theta\le\pi\}, traversed counterclockwise. Since ff is meromorphic with only the poles z1,…,znz_1,\dots,z_n inside CRC_R (by hypothesis (i), all poles in the upper half-plane are among these, and CRC_R contains no poles on its boundary by hypothesis on ff's poles avoiding R\mathbb{R} and choice of RR), the Residue Theorem gives

∮CRf(z)eiaz dz=2πi∑k=1nRes⁡(f(z)eiaz,zk).\oint_{C_R} f(z)e^{iaz}\,dz = 2\pi i \sum_{k=1}^n \operatorname{Res}(f(z)e^{iaz}, z_k).

Splitting the contour,

∫−RRf(x)eiax dx+∫ΓRf(z)eiaz dz=2πi∑k=1nRes⁡(f(z)eiaz,zk).\int_{-R}^{R} f(x)e^{iax}\,dx + \int_{\Gamma_R} f(z)e^{iaz}\,dz = 2\pi i\sum_{k=1}^n \operatorname{Res}(f(z)e^{iaz}, z_k).

Case a=0a=0. Here eiaz≡1e^{iaz}\equiv 1, and by hypothesis (iii) and the ML Inequality and Estimates,

∣∫ΓRf(z) dz∣≤MR⋅πR→0\left|\int_{\Gamma_R} f(z)\,dz\right| \le M_R \cdot \pi R \to 0

as R→∞R\to \infty, since RMR→0R M_R \to 0 by hypothesis. Hence ∫ΓRf→0\int_{\Gamma_R} f \to 0. Case a>0a>0. By hypothesis (iv), MR:=max⁡ΓR∣f∣→0M_R := \max_{\Gamma_R}|f| \to 0, so Jordan's Lemma applies directly with g=fg=f, giving ∫ΓRf(z)eiaz dz→0\int_{\Gamma_R} f(z)e^{iaz}\,dz \to 0. In either case, letting R→∞R \to \infty in the split equation, using the assumed convergence of the improper integral ∫−∞∞f(x)eiax dx\int_{-\infty}^\infty f(x)e^{iax}\,dx (so that ∫−RR→∫−∞∞\int_{-R}^R \to \int_{-\infty}^\infty) and the vanishing of the arc term, yields

∫−∞∞f(x)eiax dx=2πi∑k=1nRes⁡(f(z)eiaz,zk).\int_{-\infty}^\infty f(x)e^{iax}\,dx = 2\pi i \sum_{k=1}^n \operatorname{Res}(f(z)e^{iaz}, z_k).

The right side does not depend on RR once RR exceeds all ∣zk∣|z_k|, since the sum is over the fixed, finite set of poles in the upper half-plane; this confirms the limit is simply this constant. ■\blacksquare

∎

Worked examples

Example 1 (rational function, asymmetric denominator, semicircle + direct ML). Compute

I=∫−∞∞x2 dx(x2+4)(x2+2x+5).I = \int_{-\infty}^{\infty} \frac{x^2\,dx}{(x^2+4)(x^2+2x+5)}.

Let f(z)=z2(z2+4)(z2+2z+5)f(z) = \dfrac{z^2}{(z^2+4)(z^2+2z+5)}. Factor: z2+4=(z−2i)(z+2i)z^2+4=(z-2i)(z+2i), and z2+2z+5=(z+1)2+4=(z−(−1+2i))(z−(−1−2i))z^2+2z+5=(z+1)^2+4=(z-(-1+2i))(z-(-1-2i)). The poles in the upper half-plane are z=2iz=2i and z=−1+2iz=-1+2i; both are simple, and neither lies on R\mathbb{R}, so hypotheses (i)–(ii) hold. For the arc estimate: for ∣z∣=R|z|=R large, ∣z2+4∣≥R2−4|z^2+4|\ge R^2-4 and ∣z2+2z+5∣≥R2−2R−5|z^2+2z+5| \ge R^2 - 2R - 5 (reverse triangle inequality), so

∣f(z)∣≤R2(R2−4)(R2−2R−5)=:MR,|f(z)| \le \frac{R^2}{(R^2-4)(R^2-2R-5)} =: M_R,

and RMR=R3(R2−4)(R2−2R−5)→0R M_R = \dfrac{R^3}{(R^2-4)(R^2-2R-5)} \to 0 as R→∞R\to\infty (degree 33 numerator vs. degree 44 denominator), satisfying hypothesis (iii) with a=0a=0. Residue at z=2iz=2i: since this is a simple pole of z2/[(z−2i)(z+2i)(z2+2z+5)]z^2/[(z-2i)(z+2i)(z^2+2z+5)],

Res⁡z=2if=(2i)2(2i+2i)((2i)2+2(2i)+5)=−44i (−4+4i+5)=−44i(1+4i).\operatorname{Res}_{z=2i} f = \frac{(2i)^2}{(2i+2i)\big((2i)^2+2(2i)+5\big)} = \frac{-4}{4i\,(-4+4i+5)} = \frac{-4}{4i(1+4i)}.

Compute 4i(1+4i)=4i+16i2=−16+4i4i(1+4i) = 4i + 16i^2 = -16+4i. So

Res⁡z=2if=−4−16+4i=−1−4+i=−1(−4−i)(−4)2+12=4+i17.\operatorname{Res}_{z=2i} f = \frac{-4}{-16+4i} = \frac{-1}{-4+i} = \frac{-1(-4-i)}{(-4)^2+1^2} = \frac{4+i}{17}.

Residue at z=−1+2iz=-1+2i: write z2+2z+5=(z−(−1+2i))(z−(−1−2i))z^2+2z+5=(z-(-1+2i))(z-(-1-2i)), so

Res⁡z=−1+2if=(−1+2i)2((−1+2i)2+4)((−1+2i)−(−1−2i)).\operatorname{Res}_{z=-1+2i} f = \frac{(-1+2i)^2}{\big((-1+2i)^2+4\big)\big((-1+2i)-(-1-2i)\big)}.

First, (−1+2i)2=1−4i+4i2=1−4i−4=−3−4i(-1+2i)^2 = 1 -4i +4i^2 = 1-4i-4 = -3-4i. Then (−1+2i)2+4=1−4i(-1+2i)^2+4 = 1-4i. And (−1+2i)−(−1−2i)=4i(-1+2i)-(-1-2i)=4i. So

Res⁡z=−1+2if=−3−4i(1−4i)(4i).\operatorname{Res}_{z=-1+2i} f = \frac{-3-4i}{(1-4i)(4i)}.

Compute (1−4i)(4i)=4i−16i2=16+4i(1-4i)(4i) = 4i -16i^2 = 16+4i. So

Res⁡z=−1+2if=−3−4i16+4i=(−3−4i)(16−4i)162+42=(−48+12i−64i+16i2)272=−48−16−52i272=−64−52i272.\operatorname{Res}_{z=-1+2i} f = \frac{-3-4i}{16+4i} = \frac{(-3-4i)(16-4i)}{16^2+4^2} = \frac{(-48+12i-64i+16i^2)}{272} = \frac{-48-16-52i}{272} = \frac{-64-52i}{272}.

Simplify: −64−52i272=−16−13i68\dfrac{-64-52i}{272} = \dfrac{-16-13i}{68} (dividing numerator and denominator by 4). Sum of residues:

4+i17+−16−13i68=16+4i68+−16−13i68=−9i68.\frac{4+i}{17} + \frac{-16-13i}{68} = \frac{16+4i}{68} + \frac{-16-13i}{68} = \frac{-9i}{68}.

By the theorem,

I=2πi⋅−9i68=18π68=9π34.I = 2\pi i \cdot \frac{-9i}{68} = \frac{18\pi}{68} = \frac{9\pi}{34}.

As a sanity check, II must be real and positive (integrand is a nonnegative rational function decaying like 1/x21/x^2, clearly integrable), and indeed 9π/34≈0.8319\pi/34 \approx 0.831, a plausible positive value.

Example 2 (oscillatory integral via Jordan's Lemma, asymmetric shift). Compute

J=∫−∞∞cos⁡(2x)(x−1)2+9 dx.J = \int_{-\infty}^\infty \frac{\cos(2x)}{(x-1)^2+9}\,dx.

Write J=Re⁡(∫−∞∞e2ix(x−1)2+9 dx)J = \operatorname{Re}\left(\int_{-\infty}^\infty \frac{e^{2ix}}{(x-1)^2+9}\,dx\right), since the denominator is real and the imaginary part of the integral corresponds to ∫sin⁡(2x)/[(x−1)2+9] dx\int \sin(2x)/[(x-1)^2+9]\,dx, which is genuinely part of a different (odd-type) computation — here we only need the real part. Let f(z)=1(z−1)2+9f(z) = \dfrac{1}{(z-1)^2+9}, a=2>0a=2>0. The poles of ff solve (z−1)2=−9(z-1)^2=-9, i.e. z−1=±3iz-1=\pm 3i, so z=1+3iz = 1+3i or z=1−3iz=1-3i. Only z1=1+3iz_1 = 1+3i lies in the upper half-plane; it is a simple pole. Since ∣f(z)∣≤1R2−2R−8→0|f(z)| \le \dfrac{1}{R^2 - 2R -8}\to 0 as R→∞R\to\infty (reverse triangle inequality on ∣z−1∣≥∣z∣−1|z-1|\ge |z|-1), hypothesis (iv) holds, and hypotheses (i)–(ii) are clear (finite poles, none real, since 1±3i∉R1\pm 3i \notin \mathbb R). Compute the residue: since (z−1)2+9=(z−(1+3i))(z−(1−3i))(z-1)^2+9 = (z-(1+3i))(z-(1-3i)),

Res⁡z=1+3i[f(z)e2iz]=e2i(1+3i)(1+3i)−(1−3i)=e2i−66i=e−6e2i6i.\operatorname{Res}_{z=1+3i}\left[f(z)e^{2iz}\right] = \frac{e^{2i(1+3i)}}{(1+3i)-(1-3i)} = \frac{e^{2i - 6}}{6i} = \frac{e^{-6}e^{2i}}{6i}.

By the theorem,

∫−∞∞e2ix(x−1)2+9 dx=2πi⋅e−6e2i6i=πe−63 e2i=πe−63(cos⁡2+isin⁡2).\int_{-\infty}^\infty \frac{e^{2ix}}{(x-1)^2+9}\,dx = 2\pi i \cdot \frac{e^{-6}e^{2i}}{6i} = \frac{\pi e^{-6}}{3}\,e^{2i} = \frac{\pi e^{-6}}{3}\big(\cos 2 + i\sin 2\big).

Taking the real part,

J=πe−63cos⁡2.J = \frac{\pi e^{-6}}{3}\cos 2.

Numerically, e−6≈0.00247875e^{-6}\approx 0.00247875, cos⁡2≈−0.41615\cos 2 \approx -0.41615, so J≈3.14159×0.00247875×(−0.41615)3≈−0.001080J \approx \dfrac{3.14159\times 0.00247875 \times(-0.41615)}{3} \approx -0.001080. The negative sign is plausible: the Lorentzian weight is centered at x=1x=1 and cos⁡(2x)\cos(2x) near x=1x=1 is cos⁡2<0\cos 2 <0, and the weight decays fast enough that this local sign dominates the oscillatory average.

CounterexampleJordan's Lemma fails without the decay hypothesis MR→0M_R\to 0

Consider f(z)≡1f(z) \equiv 1 (constant), a=1a=1, and try to apply the argument underlying Jordan's Lemma to conclude

lim⁡R→∞∫ΓReiz dz=0,\lim_{R\to\infty}\int_{\Gamma_R} e^{iz}\,dz = 0,

where ΓR\Gamma_R is the upper semicircular arc of radius RR. Here MR=max⁡ΓR∣f(z)∣=1M_R = \max_{\Gamma_R}|f(z)| = 1 for all RR, so the hypothesis MR→0M_R \to 0 of Jordan's Lemma fails (it does not even hold that MR→0M_R\to0; MRM_R is constantly 11). Direct computation: parametrizing z=Reiθz=Re^{i\theta},

∫ΓReiz dz=∫0πeiReiθ iReiθ dθ.\int_{\Gamma_R} e^{iz}\,dz = \int_0^\pi e^{iRe^{i\theta}}\, iRe^{i\theta}\,d\theta.

This does not vanish as R→∞R\to \infty; in fact it does not even converge to a finite limit in general (its modulus is bounded by πR\pi R by the crude ML estimate, but a finer analysis via the same computation as in the proof of Jordan's Lemma shows only

∣∫ΓReiz dz∣≤π1⋅1=π is NOT what we get, since MR↛0,\left|\int_{\Gamma_R}e^{iz}\,dz\right| \le \frac{\pi}{1}\cdot 1 = \pi \text{ is NOT what we get, since } M_R \not\to 0,

and indeed the bound πMR/a\pi M_R/a used in the proof becomes just π\pi, a nonzero constant — no decay is forced). Concretely, one can compute this integral exactly: substitute u=Reiθu=Re^{i\theta}, so the antiderivative of eize^{iz} is −ieiz-ie^{iz}, giving

∫ΓReiz dz=[−ieiz]z=Rz=−R=−i(e−iR−eiR)=−i(−2isin⁡R)=−2sin⁡R,\int_{\Gamma_R} e^{iz}\,dz = \Big[-ie^{iz}\Big]_{z=R}^{z=-R} = -i\big(e^{-iR}-e^{iR}\big) = -i(-2i\sin R) = -2\sin R,

(here endpoints of ΓR\Gamma_R, oriented from θ=0\theta=0 i.e. z=Rz=R to θ=π\theta=\pi i.e. z=−Rz=-R, are used). This oscillates as −2sin⁡R-2\sin R and does not tend to 00 as R→∞R\to\infty; it does not even have a limit. This shows precisely how the hypothesis MR→0M_R \to 0 (decay of ff itself, not just boundedness) is essential to Jordan's Lemma, and hence to the semicircular-contour method for oscillatory integrals: a bounded, non-decaying ff can make the arc contribution persist forever.

Scope, edge cases & common traps

SituationCorrect contour / toolCommon error
Rational ff, deg⁡(denom)≥deg⁡(numer)+2\deg(\text{denom}) \ge \deg(\text{numer})+2Semicircle, plain ML estimateUsing ML when degree gap is only 11 (arc does not vanish)
f(x)eiaxf(x)e^{iax}, a>0a>0, ff merely O(1/R)O(1/R) or smallerSemicircle + Jordan's LemmaApplying plain ML (fails since MR⋅πR↛0M_R\cdot \pi R \not\to 0) instead of Jordan's sharper bound
f(x)e−iaxf(x)e^{-iax}, a>0a>0Lower half-plane semicircle (clockwise), poles thereForgetting to flip contour orientation, picking wrong poles
Integrand with branch point, e.g. xα/(1+x)x^\alpha/(1+x)Keyhole contour with branch cut avoidedIgnoring monodromy jump across the cut, mis-assigning the branch of zαz^\alpha
Simple pole exactly on R\mathbb{R}Indent with small semicircle, small-arc lemma gives ∓iπRes⁡\mp i\pi\operatorname{Res}Treating the principal value integral as if the pole were not there at all
∫02πR(cos⁡θ,sin⁡θ) dθ\int_0^{2\pi} R(\cos\theta,\sin\theta)\,d\thetaUnit circle, z=eiθz=e^{i\theta} substitutionForgetting the Jacobian factor dz=iz dθdz = iz\,d\theta, or double counting poles at z=0z=0

The mathematician behind it

Origins · 1 minAugustin-Louis Cauchy and how this was discovered
AC
Augustin-Louis Cauchy
1789–1857 · Founder of complex function theory, École Polytechnique and Sorbonne, Paris

Cauchy built the theory of contour integration and residues that makes this entire toolkit possible: the Cauchy Integral Theorem, the Cauchy Integral Formula, and — building on work later completed by others — the residue calculus itself. He was notoriously prolific (over 800 papers) and used complex-variable methods to attack real definite integrals that had stumped Euler and Laplace using purely real techniques, turning what were isolated tricks into a systematic method. Camille Jordan, whose lemma bounds the semicircular arc for oscillatory integrands, worked a generation later, consolidating rigorous real analysis (uniform convergence, careful ϵ\epsilon–δ\delta estimates) that put Cauchy's formal manipulations on solid ground.

History
- 1814 — Cauchy submits his first memoir on definite integrals via complex substitution, prefiguring contour methods.
- 1825 — Cauchy proves what is now the Cauchy Integral Theorem for integrals of analytic functions around closed contours.
- 1831 — Cauchy states the Cauchy Integral Formula, from which the residue calculus is later organized.
- 1846 — Cauchy formalizes the notion of residue and articulates an early residue theorem for evaluating definite integrals.
- 1893 — Camille Jordan's Cours d'Analyse popularizes the rigorous arc-vanishing estimate now called Jordan's Lemma.
Why does closing the contour in the upper half-plane (rather than the lower) matter for ∫f(x)eiax dx\int f(x)e^{iax}\,dx with a>0a>0?

The factor eiaz=eiaxe−aye^{iaz} = e^{iax}e^{-ay} (writing z=x+iyz=x+iy) decays exponentially as y→+∞y\to+\infty precisely when a>0a>0, so it is small on a large semicircle in the upper half-plane. Closing there lets Jordan's Lemma kill the arc contribution. If you instead close in the lower half-plane, eiaze^{iaz} grows exponentially as y→−∞y\to-\infty, and the arc estimate fails entirely — the arc integral typically blows up rather than vanishing. The rule of thumb "match the sign of aa to the half-plane where eiaze^{iaz} decays" is the single most common bookkeeping point to get right in this method.

What if the rational function has a pole exactly on the real axis, like ∫−∞∞sin⁡xxdx\int_{-\infty}^\infty \frac{\sin x}{x}dx?

You cannot integrate straight through a pole on the contour, since ff is not even defined there and the ordinary real integral is only understood in a principal-value sense. The fix is to indent the contour with a small semicircular detour of radius ε\varepsilon around the pole, either above or below it. The small-arc lemma shows that as ε→0\varepsilon \to 0, this small arc contributes exactly ∓iπRes⁡(f,z0)\mp i\pi \operatorname{Res}(f, z_0) for a simple pole z0z_0 (sign depending on orientation), rather than 00 or the full 2πiRes⁡2\pi i \operatorname{Res}. This partial-residue contribution is then combined with the large-arc estimate and the residues strictly inside to recover the principal value of the original integral.

Can this method handle integrals over [0,∞)[0,\infty) rather than the whole real line?

Yes, in two main ways. If the integrand is even, ∫0∞f(x) dx=12∫−∞∞f(x) dx\int_0^\infty f(x)\,dx = \tfrac12\int_{-\infty}^\infty f(x)\,dx, reducing to the semicircle method directly. If the integrand is not even — for instance it contains xαx^\alpha with non-integer α\alpha, or log⁡x\log x — one typically uses a keyhole contour: a large circle and a small circle around the origin, connected by two segments just above and just below the positive real axis, chosen because zαz^\alpha or log⁡z\log z has a branch point at 00. The two nearly-coincident segments do not cancel because the branch of the multivalued function differs by a fixed phase factor between them, and solving the resulting linear equation for the unknown integral is the key algebraic step.

Why is the ML inequality not enough for oscillatory integrands like ∫cos⁡(ax)f(x) dx\int \cos(ax)f(x)\,dx when ff only decays like 1/∣z∣1/|z|?

The plain ML estimate bounds the arc integral by MR⋅(arc length)=MR⋅πRM_R \cdot (\text{arc length}) = M_R \cdot \pi R. If MR=O(1/R)M_R = O(1/R), this product is O(1)O(1) and does not go to zero — the estimate is too crude to conclude vanishing. Jordan's Lemma exploits the extra structure of eiaze^{iaz}: on the upper semicircle, ∣eiaz∣=e−aRsin⁡θ|e^{iaz}| = e^{-aR\sin\theta}, which is exponentially small except very near the real axis (θ\theta near 00 or π\pi), and a careful integral of this exponential factor over θ\theta produces an extra factor of 1/R1/R compared to the crude bound, exactly compensating the arc length πR\pi R. This is why Jordan's Lemma, not the plain ML inequality, is the correct and necessary tool for Fourier-type integrands.

Does the choice of contour radius RR matter, or could different large RR give different final answers?

No — once RR is large enough to enclose every pole of interest inside the upper (or lower) half-plane, the Residue Theorem gives exactly 2πi2\pi i times the sum over that fixed, finite set of poles, independent of RR. Increasing RR further does not add new poles (there are only finitely many), so the residue sum on the right-hand side is a constant once RR passes the threshold max⁡k∣zk∣\max_k|z_k|. The only role of letting R→∞R\to\infty is to make the diameter segment converge to the full improper integral and to make the arc's contribution vanish (or converge to its known limit); the residue sum itself is settled the moment all poles are enclosed.

What goes wrong if we forget to check that the poles are not exactly on the chosen contour?

The Residue Theorem, in its basic form, requires the contour to avoid all singularities of the function — the theorem simply does not apply, and none of its conclusions (finite value, computable via residues) are guaranteed, if a pole sits on the contour itself, since ff is undefined or the contour integral is not even classically defined there. Concretely the real integral becomes an improper (possibly principal-value) integral needing separate justification, and one must either indent the contour with a small arc (using the small-arc lemma) or interpret the original real integral in the principal-value sense from the start, matching that convention consistently on both sides of the equation.

Is this method limited to integrands built from elementary functions like polynomials, exponentials, and trig functions?

The core requirement is only that the integrand extends to a function that is meromorphic (or meromorphic after introducing a branch cut) on a suitable region of C\mathbb{C}, with finitely many singularities enclosed by the chosen contour and controllable behavior at infinity or near branch points. This covers a vast class: rational functions, rational combinations of sin⁡,cos⁡,eiax\sin,\cos,e^{iax}, functions with algebraic branch points (xαx^\alpha), and logarithms. It does not directly apply to integrands with no natural analytic continuation, essential singularities accumulating on the contour, or unavoidably infinitely many poles inside every contour (though summation techniques exist for some such cases, they go beyond the scope of the basic ML/Jordan/keyhole/indentation toolkit).

Before this