Cauchy's Integral Formula and Derivatives

Cauchy's integral theorem is a statement about nothing: the loop integral of an analytic function vanishes. Cauchy's integral formula is what happens when you deliberately spoil analyticity at a single interior point. Divide the integrand by $(z-a)$ , planting an isolated pole at $a$ , and the loop integral stops vanishing — instead it returns the exact number $f(a)$ . One puncture converts a null result into a machine that reads off the value of $f$ at any point you choose, using only its values along a distant boundary. That is the whole idea, and everything downstream is bookkeeping on top of it. The formula wears two faces. The first reproduces $f(a)$ from boundary data. The second, obtained by differentiating the integrand with respect to the parameter $a$ under the integral sign, delivers every derivative $f^{(n)}(a)$ as its own contour integral. The corollary is one of the most startling facts in mathematics: a function differentiable a single time in the complex sense is automatically differentiable infinitely often, and equals its own convergent Taylor series. The entire hierarchy of real-variable smoothness classes — $C^1$ , $C^2$ , $C^\infty$ , analytic — collapses to a single rung. Read physically, the kernel $1/[2\pi i(z-a)]$ is a reproducing Green's function that samples $f$ at $a$ . It converts the local datum "value at a point" into a global integral over a boundary, exactly the logic of potential theory, two-dimensional electrostatics, and dispersion relations. From this one identity flow Cauchy's inequalities, Liouville's theorem, the maximum-modulus principle, Morera's converse, the Taylor and Laurent expansions, and finally the residue calculus that physicists use to evaluate integrals no real-variable method can touch. Master this formula and the rest of contour integration is downhill.

Key takeaways

Cauchy's integral formula says an analytic function inside a simple closed contour is fixed entirely by its boundary values: $f(a)=\frac{1}{2\pi i}\oint_C \frac{f(z)}{z-a}\,\mathrm{d}z$ .
Differentiating under the integral sign gives every derivative, $f^{(n)}(a)=\frac{n!}{2\pi i}\oint_C \frac{f(z)}{(z-a)^{n+1}}\,\mathrm{d}z$ — so one complex derivative forces infinitely many.
It follows from Cauchy's theorem by shrinking $C$ to a tiny circle about $a$ and using the single elementary fact $\oint \mathrm{d}z/(z-a)=2\pi i$ .
Cauchy's inequalities $|f^{(n)}(a)|\le n!\,M/R^n$ bound derivatives by boundary data and instantly yield Liouville's theorem: a bounded entire function is constant.
It is the computational engine behind Taylor and Laurent series, the mean-value and maximum-modulus principles, and the residue theorem used throughout physics.
Key idea: inserting a simple pole at an interior point $a$ turns Cauchy's theorem into a sampling device — the contour integral of $f(z)/(z-a)^{n+1}$ reproduces $f$ and all its derivatives at $a$ from boundary values alone.

Where this lives today

The integral formula is not a museum piece; it is running inside working code and instruments right now. Its most direct modern descendant is the Cauchy method of numerical differentiation and integration: to compute a high-order derivative $f^{(n)}(a)$ without the catastrophic cancellation of finite differences, one samples $f$ at points on a circle $z_k=a+Re^{2\pi ik/N}$ and forms $f^{(n)}(a)\approx \frac{n!}{R^n N}\sum_{k=0}^{N-1} f(z_k)\,e^{-2\pi ink/N}$ — a discretised contour integral evaluated by the FFT, stable to machine precision. The same idea, applied to $\log f$ , is the argument-principle root finder: winding a contour and counting the change in phase locates zeros and poles of analytic functions and transfer functions, which is how control engineers and photonics designers hunt resonances in the complex-frequency plane. In many-body and statistical physics the formula underlies spectral (Matsubara) sums, where a contour is deformed to convert an intractable sum over frequencies into a residue calculation, and in scattering theory the analytic continuation of amplitudes and the Kramers-Kronig dispersion relations are Cauchy's formula wearing a physicist's hat — the real and imaginary parts of a causal response function are tied together precisely because the response is the boundary value of an analytic function. Even the maximum-modulus principle it implies is a daily tool, bounding transfer functions in robust-control $H^\infty$ design. Wherever a physicist writes "close the contour in the upper half-plane," this formula is doing the work.

The physical picture

Picture the kernel $1/[2\pi i(z-a)]$ laid over the interior of a loop. Away from $a$ it is a gentle, analytic weight, and integrating $f$ against it around a closed curve would give zero by Cauchy's theorem. But at $z=a$ the kernel blows up: it is an infinitely sharp spike sitting on the single interior point. Integrating $f$ against that spike does not average $f$ — it samples it, extracting the one number $f(a)$ and discarding everything else. Because the theorem lets you deform the loop freely through the region where the integrand is analytic, you may collapse any enclosing contour down to a hair-thin circle hugging $a$ , where only the local value of $f$ survives. The first figure shows this collapse and the one elementary integral, $\oint \mathrm{d}z/(z-a)=2\pi i$ , that powers it.

Figure

The reproducing kernel 1/(2πi(z−a)) samples f at the interior point a: any enclosing loop C deforms down to a tiny circle C_ρ without changing the integral, because ∮ dz/(z−a) = 2πi on every circle around a, so the loop returns exactly f(a)

The derivative version is the same trick applied to a fatter pole. Differentiating the kernel $n$ times with respect to $a$ turns $1/(z-a)$ into $n!/(z-a)^{n+1}$ — a spike so sharp it samples not the value of $f$ but its $n$ -th derivative. The factorial that appears is the very same $n!$ that sits under the Taylor coefficient $f^{(n)}(a)/n!$ , which is no accident: the integral formula and the Taylor series are two readings of one identity. The second figure makes the payoff concrete with a real number. A pole placed inside the contour lifts out the value there; a point outside contributes nothing, because there the integrand is analytic and Cauchy's theorem reasserts itself.

Figure

A pole placed at z = 2 inside the contour makes ∮_C e^z/(z−2) dz = 2πi·e² ≈ 46.4 i, while an exterior point such as z = 5 contributes nothing (∮ = 0) because the integrand is analytic there — inside samples, outside vanishes

The core equations

The basic formula is the heart of the subject. It asserts that the value of an analytic function at an interior point is a weighted integral of its boundary values, with the Cauchy kernel supplying the weight. Read it as a reproducing identity, not merely a way to compute an integral.

f(a)=\frac{1}{2\pi i}\oint_C \frac{f(z)}{z-a}\,\mathrm{d}z \qquad (1)

where: $f$ — a function analytic on and inside the simple closed contour $C$ ; $a$ — a point strictly interior to $C$ , where $f$ is evaluated; $C$ — a positively oriented (counterclockwise) simple closed contour; $z$ — the integration variable running along $C$ . Reading: the boundary data $\{f(z):z\in C\}$ over-determine $f$ everywhere inside, so an analytic function has no interior freedom once its boundary is fixed. Differentiating (1) under the integral sign with respect to the parameter $a$ — legal because $a$ stays a fixed distance from $C$ — reproduces every derivative. Each differentiation lowers the kernel's power by one and pulls out a factor, leaving the factorial.

f^{(n)}(a)=\frac{n!}{2\pi i}\oint_C \frac{f(z)}{(z-a)^{n+1}}\,\mathrm{d}z \qquad (2)

where: $f^{(n)}$ — the $n$ -th complex derivative of $f$ at $a$ ; $n!$ — the factorial produced by differentiating $1/(z-a)$ a total of $n$ times; $(z-a)^{n+1}$ — the denominator after those $n$ differentiations. Reading: because the right side exists for every $n$ the instant $f$ is continuous on $C$ , analyticity once implies analyticity infinitely often — the collapse of smoothness classes with no real-variable analogue. Specialising $C$ to a circle of radius $R$ centred on $a$ and writing $z=a+Re^{i\theta}$ turns (1) into an average. The Cauchy kernel becomes constant on the circle and the integral reduces to a plain angular mean.

f(a)=\frac{1}{2\pi}\int_0^{2\pi} f(a+Re^{i\theta})\,\mathrm{d}\theta \qquad (3)

where: $R$ — the radius of a circle about $a$ lying within the domain of analyticity; $\theta$ — the angular parameter along that circle. Reading: the value at the centre equals the mean of the values around any surrounding circle — the complex-analytic mean-value property, whose real part is the mean-value theorem for harmonic potentials. Applying the ML (length-times-max) estimate to the derivative formula (2) on that same circle bounds every derivative by the maximum modulus of $f$ on the boundary. These are Cauchy's inequalities, the quantitative core of the theory's rigidity.

|f^{(n)}(a)| \le \frac{n!\,M(R)}{R^n},\qquad M(R)=\max_{|z-a|=R}|f(z)| \qquad (4)

where: $M(R)$ — the maximum modulus of $f$ on the circle $|z-a|=R$ ; $R$ — the radius of that circle within the domain. Reading: derivatives cannot be larger than the boundary data allow; sending $R\to\infty$ for a bounded entire function kills every derivative and proves Liouville's theorem. If the contour is allowed to wind around $a$ more than once, the formula picks up the signed number of loops. This is the topological generalisation and the seed of the argument principle.

\frac{1}{2\pi i}\oint_C \frac{f(z)}{z-a}\,\mathrm{d}z = n(C,a)\,f(a) \qquad (5)

where: $n(C,a)$ — the winding number (index) of $C$ about $a$ , an integer counting signed encirclements. Reading: a simple counterclockwise loop has $n(C,a)=1$ and recovers (1); a doubly wound loop returns $2f(a)$ ; an exterior point has $n(C,a)=0$ and the integral vanishes. Finally, the derivative formula packaged as a coefficient extractor is the residue at a pole of order $n$ — the bridge from this article to the residue theorem.

\operatorname{Res}_{z=a} g = \frac{1}{(n-1)!}\lim_{z\to a}\frac{\mathrm{d}^{\,n-1}}{\mathrm{d}z^{\,n-1}}\big[(z-a)^n g(z)\big] \qquad (6)

where: $g=f/(z-a)^n$ with $f$ analytic near $a$ — a function with a pole of order $n$ ; $\operatorname{Res}$ — the coefficient of $(z-a)^{-1}$ in the Laurent expansion. Reading: the residue is just $f^{(n-1)}(a)/(n-1)!$ read off through (2), which is why the residue theorem is the integral formula wearing a Laurent-series costume.

Derivation AProof of the basic Cauchy integral formula

The goal is to extract the single number $f(a)$ from a loop integral by two moves that are individually elementary: deform the contour down to a vanishingly small circle around $a$ using Cauchy's theorem, then split the integrand into a piece whose integral is the exact constant $2\pi i$ and a remainder that the continuity of $f$ forces to zero. Keep track only of the fact that $f$ is analytic on and inside a positively oriented simple closed contour $C$ with $a$ strictly interior, and let $I=\oint_C f(z)/(z-a)\,\mathrm{d}z$ be the quantity in question.

Step 1. Deform to a small circle. The integrand $f(z)/(z-a)$ is analytic everywhere in the annular region between $C$ and a small circle $C_\rho:\,|z-a|=\rho$ , so Cauchy's theorem (in its deformation form) gives $I=\oint_{C_\rho} f(z)/(z-a)\,\mathrm{d}z$ . Why: shrinking the loop costs nothing wherever the integrand is analytic, and it concentrates all the action onto a circle where $z-a$ has constant modulus $\rho$ .

Step 2. Add and subtract the target value: $I=\oint_{C_\rho}\frac{f(z)-f(a)}{z-a}\,\mathrm{d}z + f(a)\oint_{C_\rho}\frac{\mathrm{d}z}{z-a}$ . Why: this isolates the constant $f(a)$ , which we want to survive, from a numerator $f(z)-f(a)$ that we will show is negligible.

Step 3. Evaluate the kernel integral. Parametrise $z=a+\rho e^{i\theta}$ , $\mathrm{d}z=i\rho e^{i\theta}\,\mathrm{d}\theta$ , so $\oint_{C_\rho}\mathrm{d}z/(z-a)=\int_0^{2\pi} i\rho e^{i\theta}/(\rho e^{i\theta})\,\mathrm{d}\theta=\int_0^{2\pi} i\,\mathrm{d}\theta=2\pi i$ , independent of $\rho$ . Why: this one elementary integral is the entire engine — the factor $2\pi i$ that appears everywhere in contour integration is born here.

Step 4. Bound the remainder. By the ML estimate, $\big|\oint_{C_\rho}\frac{f(z)-f(a)}{z-a}\,\mathrm{d}z\big|\le \frac{\max_{C_\rho}|f(z)-f(a)|}{\rho}\cdot 2\pi\rho = 2\pi\,\max_{C_\rho}|f(z)-f(a)|$ . Continuity of $f$ at $a$ makes $\max_{C_\rho}|f(z)-f(a)|\to 0$ as $\rho\to 0$ , so the remainder vanishes. Why: the length $2\pi\rho$ and the denominator $\rho$ cancel exactly, leaving only the modulus of continuity, which continuity kills.

Step 5. Take the limit. Since $I$ does not depend on $\rho$ , letting $\rho\to 0$ gives $I=0+2\pi i\,f(a)$ ; dividing by $2\pi i$ yields $f(a)=\frac{1}{2\pi i}\oint_C f(z)/(z-a)\,\mathrm{d}z$ . Why: the left side was fixed all along, so the vanishing of the remainder pins the whole integral to the surviving constant term.

Go deeper · Derivation B · 1 minThe derivative formula by differentiating under the integral

Derivation BThe derivative formula by differentiating under the integral

The goal is to promote the basic formula to a representation of every derivative, and in doing so to prove the signature theorem of complex analysis — that one complex derivative implies infinitely many. The cleanest route forms the difference quotient of the basic formula, simplifies the algebra of two nearby kernels, and passes to the limit under the integral sign, which is justified because $a$ sits a fixed positive distance $d$ from the contour $C$ so the integrand never becomes singular. Iterating the same move generates the factorial and raises the power.

Step 1. Form the difference quotient from formula (1) evaluated at $a+h$ and at $a$ : $\frac{f(a+h)-f(a)}{h}=\frac{1}{2\pi i}\oint_C f(z)\,\frac{1}{h}\Big[\frac{1}{z-a-h}-\frac{1}{z-a}\Big]\mathrm{d}z$ . Why: derivatives are limits of difference quotients, and here the only $h$ -dependence sits harmlessly inside the kernel, not in $f$ .

Step 2. Simplify the bracket over a common denominator: $\frac{1}{h}\big[\frac{1}{z-a-h}-\frac{1}{z-a}\big]=\frac{1}{(z-a-h)(z-a)}$ , the $h$ in the numerator cancelling exactly. Why: the cancellation removes the apparent $0/0$ and leaves a kernel that is manifestly finite for $z$ on $C$ .

Step 3. Let $h\to 0$ . For $a$ at distance $d>0$ from $C$ and $|h|<d/2$ the integrand is uniformly bounded, so the limit passes inside the integral: $f'(a)=\frac{1}{2\pi i}\oint_C \frac{f(z)}{(z-a)^2}\,\mathrm{d}z$ . Why: uniform boundedness on the compact contour licenses the dominated passage to the limit, turning a difference quotient into an honest derivative.

Step 4. Iterate. Repeating the difference-quotient argument on $f'$ , then $f''$ , and so on multiplies the surviving factor at each stage, giving $f^{(n)}(a)=\frac{n!}{2\pi i}\oint_C \frac{f(z)}{(z-a)^{n+1}}\,\mathrm{d}z$ . Because the right side exists for every $n$ as soon as $f$ is merely continuous on $C$ , analyticity once forces all derivatives to exist. Why: nothing in the construction ever demanded more smoothness of $f$ than a single complex derivative — the integral supplies the rest for free.

Go deeper · Derivation C · 1 minCauchy's inequalities and Liouville's theorem

Derivation CCauchy's inequalities and Liouville's theorem

The goal is to convert the derivative representation into a hard numerical bound and then extract from it the most famous rigidity result in the subject — that a bounded function analytic on the whole plane must be constant. The strategy is to apply the ML estimate to formula (2) on a circle of radius $R$ , read off Cauchy's inequalities, then specialise to an entire function and let the radius run to infinity, where the bound crushes the first derivative to zero at every point.

Step 1. Estimate the derivative integral on the circle $|z-a|=R$ . With $|f(z)|\le M(R)$ , denominator $|z-a|^{n+1}=R^{n+1}$ , and contour length $2\pi R$ , the ML bound gives $|f^{(n)}(a)|\le \frac{n!}{2\pi}\cdot\frac{M(R)}{R^{n+1}}\cdot 2\pi R$ . Why: the length-times-max rule turns an integral we cannot compute into an inequality we can, using only the size of $f$ on the boundary.

Step 2. Simplify to Cauchy's inequalities: $|f^{(n)}(a)|\le \frac{n!\,M(R)}{R^n}$ . Why: this is the quantitative statement that derivatives are controlled by boundary data — the analytic version of "you cannot wiggle the interior without paying at the boundary".

Step 3. Specialise to an entire, bounded function. If $|f|\le M$ everywhere then $M(R)\le M$ for all $R$ , and taking $n=1$ gives $|f'(a)|\le M/R$ . Why: entireness means the circle can be as large as we like, and boundedness means the numerator never grows to compensate.

Step 4. Let $R\to\infty$ : $|f'(a)|\le M/R\to 0$ , so $f'(a)=0$ for every $a$ , hence $f$ is constant. This is Liouville's theorem, and it delivers the fundamental theorem of algebra as an immediate corollary (a nonconstant polynomial with no root would make $1/p$ a bounded entire function). Why: a bound that shrinks to zero as the radius grows leaves no room for any variation at all — global boundedness plus global analyticity is enough to freeze the function.

Limiting cases

The formula specialises cleanly across the regimes that matter in practice:

Case	Result	Meaning
Point outside the contour	$\oint_C \frac{f(z)}{z-a}\,\mathrm{d}z = 0,\quad a\notin\operatorname{int}(C)$	With $a$ exterior, $f(z)/(z-a)$ is analytic inside $C$ , so Cauchy's theorem gives zero — the loop is blind to it.
Circular contour (mean value)	$f(a)=\frac{1}{2\pi}\int_0^{2\pi} f(a+Re^{i\theta})\,\mathrm{d}\theta$	The centre value equals the angular average over any surrounding circle; its real part is the harmonic mean-value property.
Derivative bound	$\|f^{(n)}(a)\|\le \frac{n!\,M}{R^n},\ M=\max_{\|z-a\|=R}\|f(z)\|$	Boundary data cap every derivative; the engine of Liouville and the maximum-modulus principle.
Winding number $m$	$\frac{1}{2\pi i}\oint_C \frac{f(z)}{z-a}\,\mathrm{d}z = m\,f(a)$	A contour circling $a$ a net $m$ times scales the answer by the integer $m$ ; simple loop gives $m=1$ .
Higher-order pole	$\oint_C \frac{f(z)}{(z-a)^{n+1}}\,\mathrm{d}z=\frac{2\pi i}{n!}f^{(n)}(a)$	A pole of order $n+1$ samples the $n$ -th derivative rather than the value itself.

Common trapThe formula picks out f(a) no matter where a sits, even outside the loop

It does not. The single most common error is to apply $f(a)=\frac{1}{2\pi i}\oint_C f(z)/(z-a)\,\mathrm{d}z$ without checking where $a$ lives relative to $C$ . Three cases are genuinely different. If $a$ is strictly interior, the formula holds as written. If $a$ is exterior, the integrand is analytic throughout the interior and the integral is $0$ by Cauchy's theorem — not $f(a)$ . If $a$ lies on the contour, the integral is singular and the plain formula does not apply (one must use a principal value and a half-residue). Two further traps: the winding number must be $1$ for the bare formula — a contour that loops twice returns $2f(a)$ , not $f(a)$ — and $f$ must be analytic on and inside all of $C$ , so a second, unnoticed singularity of $f$ within the loop invalidates the naive answer and demands the full residue theorem. Always locate every singularity of the integrand and count how many the contour encloses before writing down a value.

Worked examples

1 · A simple pole inside the contour (fully numerical)

Evaluate $\displaystyle\oint_C \frac{e^z}{z-2}\,\mathrm{d}z$ , where $C$ is the circle $|z|=3$ traversed once counterclockwise.

Step 1. Locate the singularity and check enclosure. The integrand has a single simple pole at $z=2$ . Since $|2|=2<3$ , the pole lies inside $C$ , and $e^z$ is entire, so the hypotheses of the integral formula are met with $f(z)=e^z$ and $a=2$ .

Step 2. Apply formula (1) in the form $\oint_C \frac{f(z)}{z-a}\,\mathrm{d}z = 2\pi i\,f(a)$ . Here $f(a)=e^2$ , so the integral is $2\pi i\,e^2$ .

Step 3. Put in numbers: $e^2=7.389056$ , so $2\pi\,e^2 = 46.427$ and the integral equals $\mathbf{2\pi i\,e^2 \approx 46.43\,i}$ . The result is purely imaginary, as every such single-pole loop integral of a real-on-the-axis function must be.

Step 4. Sanity check with the exterior case. Had $C$ been $|z|=1$ instead, the pole $z=2$ would lie outside, the integrand would be analytic inside, and the integral would be $\mathbf{0}$ — a one-line confirmation that the answer is controlled entirely by enclosure.

2 · A double pole and the derivative formula

Evaluate $\displaystyle\oint_C \frac{z^3}{(z-1)^2}\,\mathrm{d}z$ , where $C$ is any positively oriented contour enclosing $z=1$ .

Step 1. Identify the order. The denominator $(z-1)^2$ is a pole of order $2$ , so this is the $n=1$ case of the derivative formula (2) with $f(z)=z^3$ , $a=1$ : $\oint_C \frac{f(z)}{(z-a)^{2}}\,\mathrm{d}z = 2\pi i\,f'(a)$ .

Step 2. Differentiate: $f'(z)=3z^2$ , so $f'(1)=3$ .

Step 3. Assemble: the integral equals $2\pi i\cdot 3 = \mathbf{6\pi i}$ . A naive attempt to treat the double pole as if it were simple (reading off $f(1)=1$ and writing $2\pi i$ ) gives the wrong answer — the order of the pole selects which derivative is sampled.

3 · A third-order pole with a symbolic answer

Evaluate $\displaystyle\oint_C \frac{\cos z}{(z-a)^3}\,\mathrm{d}z$ for a contour enclosing $z=a$ , then read off the special case $a=0$ .

Step 1. This is the $n=2$ derivative formula: $\oint_C \frac{f(z)}{(z-a)^{3}}\,\mathrm{d}z = \frac{2\pi i}{2!}\,f''(a)$ with $f(z)=\cos z$ .

Step 2. Differentiate twice: $f'(z)=-\sin z$ , $f''(z)=-\cos z$ , so $f''(a)=-\cos a$ .

Step 3. Assemble: the integral equals $\frac{2\pi i}{2}(-\cos a) = \mathbf{-\pi i\cos a}$ . Setting $a=0$ gives the clean number $\mathbf{-\pi i}$ , and setting $a=\pi/2$ gives $0$ because $\cos(\pi/2)=0$ — the loop can return zero even with the pole inside, when the sampled derivative happens to vanish.

Physics of the portrait

Cauchy is where rigour entered analysis. Before him the "integral around a loop" was a formal manipulation; after him it became a theorem with hypotheses, an estimate, and a proof — and out of that discipline fell a formula that reads a function's interior off its boundary.

Origins · 1 minAugustin-Louis Cauchy and how this was discovered

Augustin-Louis Cauchy

1789–1857 · complex analysis and the calculus of residues

Augustin-Louis Cauchy was the most prolific and the most controversial analyst of his century, author of some eight hundred papers and the man who first insisted that the concepts of the calculus — limit, continuity, convergence, the integral — be defined with precision rather than waved at. Trained as an engineer and briefly employed on the harbour works at Cherbourg under Napoleon, he turned to pure mathematics and, in an 1814 memoir on definite integrals, took the decisive step of letting the variable of integration wander into the complex plane. Over the next two decades he built the theory that bears his name: the integral theorem that a loop integral of an analytic function vanishes (1825), the integral formula that a punctured loop integral reproduces the function (developed through his Turin memoirs around 1831), and the "calcul des limites" — his derivative inequalities — from which the whole rigidity of analytic functions follows. A devout royalist, he followed the deposed Bourbons into exile after the 1830 revolution, teaching in Turin and Prague before returning to Paris; his refusal to swear allegiance to successive regimes cost him chairs but never his output. Contemporaries found him difficult — he was accused of losing the young Abel's and Galois's manuscripts — yet the machinery he assembled, the residue calculus above all, is the everyday tool by which physicists evaluate integrals, invert Fourier and Laplace transforms, and write down dispersion relations. What began as one man's demand for rigour is now the reflex of every working theorist.

History

1814

— Cauchy submits his memoir on definite integrals taken between imaginary limits

1825

— Cauchy proves the integral theorem that a loop integral of an analytic function vanishes

1831

— Cauchy obtains the integral formula and his derivative inequalities during his Turin period

1844

— Liouville's theorem, that a bounded entire function is constant, is established from Cauchy's bounds

1851

— Riemann's dissertation refounds analytic function theory on geometric and topological ideas

Going further

Texts. Stone & Goldbart, Mathematics for Physics, Ch. 17 (Complex Analysis I) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 11 · Hassani, Mathematical Physics, 2nd ed., Ch. 11 (Calculus of Residues) · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 6 · Ahlfors, Complex Analysis, 3rd ed., Ch. 4 · Whittaker & Watson, A Course of Modern Analysis, Ch. 5.

Provenance. Every equation here is a standard result of complex analysis, grounded in the migrated source and re-verified. A checker should independently recompute: (1) the elementary kernel integral $\oint_{|z-a|=\rho}\mathrm{d}z/(z-a)=2\pi i$ by direct parametrisation; (2) the difference-quotient passage to $f'(a)=\frac{1}{2\pi i}\oint_C f(z)/(z-a)^2\,\mathrm{d}z$ ; (3) the ML estimate giving Cauchy's inequalities and the $R\to\infty$ step of Liouville; (4) the worked numbers $2\pi i\,e^2\approx 46.43\,i$ , $6\pi i$ , and $-\pi i\cos a$ . Cross-check against Stone & Goldbart Ch. 17, Arfken-Weber-Harris Ch. 11, and Ahlfors Ch. 4.

Prerequisites. Cauchy's Integral Theorem · Complex Integration and Contour Integrals · Analytic Functions and the Cauchy-Riemann Equations · Functions of a Complex Variable and Limits. Leads to. Taylor Series of Analytic Functions · Liouville, Maximum Modulus and Morera · Laurent Series · Residues and the Residue Theorem.

Ask the tutor: Ask me to close a real-axis integral by contour, evaluate any loop integral with poles of arbitrary order, or show step by step how Cauchy's inequalities give Liouville's theorem and the fundamental theorem of algebra.

Why does dividing by (z − a) turn a zero into f(a)?

Cauchy's theorem says the loop integral of anything analytic inside is zero, so an analytic integrand carries no information about interior points. Dividing by $(z-a)$ deliberately breaks analyticity at the single point $a$ , planting a simple pole there. Now the integrand is analytic everywhere inside except at $a$ , so the loop integral no longer vanishes — it collects exactly the contribution of that one puncture. Deforming the contour to a tiny circle around $a$ shows the collected amount is $2\pi i$ times the local value $f(a)$ . The pole acts as a probe: it is the mechanism by which a global boundary integral is made to report a single interior value.

How can one complex derivative possibly imply infinitely many?

In real analysis a function can be differentiable once and no more — the derivative can be jagged. The complex case is different because complex differentiability is far more restrictive: it forces the Cauchy-Riemann equations, hence analyticity. The integral formula makes the consequence concrete. Once $f$ is analytic on and inside $C$ , formula (2) writes each derivative $f^{(n)}(a)$ as a contour integral of $f$ divided by $(z-a)^{n+1}$ . That integral plainly exists for every $n$ as soon as $f$ is merely continuous on the contour — no further smoothness of $f$ is ever demanded. So all derivatives exist automatically, and $f$ equals its own Taylor series. The rigidity has no real-variable analogue.

What is the physical meaning of the factor 2πi?

It is the value of the one integral that everything rests on: $\oint \mathrm{d}z/(z-a)=2\pi i$ around any circle enclosing $a$ . Parametrising $z=a+\rho e^{i\theta}$ turns the integrand into $i\,\mathrm{d}\theta$ , and integrating $\theta$ once around from $0$ to $2\pi$ gives $2\pi i$ . The $2\pi$ is the total angle swept; the $i$ records that going around in the complex plane rotates phase. Physically it is a statement of winding: the logarithm $\log(z-a)$ increases its imaginary part by $2\pi$ each time you loop the branch point once, and the integral measures that phase accumulation. Every $2\pi i$ you meet in residue calculus traces back to this single loop.

When does the formula give zero instead of f(a)?

Whenever the contour does not enclose the point in question with winding number one. If $a$ lies outside $C$ , the integrand $f(z)/(z-a)$ is analytic throughout the interior, so Cauchy's theorem returns exactly zero. Zero can also appear even with the pole enclosed — if the sampled quantity happens to vanish. In example 3 above, $\oint \cos z/(z-a)^3\,\mathrm{d}z=-\pi i\cos a$ , which is zero at $a=\pi/2$ because the second derivative $-\cos(\pi/2)=0$ . So a vanishing loop integral means one of two things: the singularity is not enclosed, or it is enclosed but the particular derivative it samples is zero. Always check which.

How is this formula the seed of the residue theorem?

The residue theorem says a loop integral equals $2\pi i$ times the sum of residues of the enclosed singularities. Cauchy's integral formula is that statement for the simplest case — a single simple pole, whose residue is $f(a)$ . The derivative formula extends it to poles of any order: a pole of order $n$ at $a$ has residue $f^{(n-1)}(a)/(n-1)!$ , which is exactly what formula (2) computes. Laurent's theorem then guarantees that near any isolated singularity the function has a series whose $(z-a)^{-1}$ coefficient is the residue. Adding up several enclosed singularities and applying the formula to each gives the full residue theorem. In short, the residue theorem is Cauchy's integral formula summed over poles.

Does the shape of the contour matter?

Only its topology matters, never its exact shape. As long as $f$ is analytic on and inside the contour and $a$ is enclosed once counterclockwise, you may stretch, dent, or balloon $C$ freely and the integral is unchanged — that is the deformation invariance guaranteed by Cauchy's theorem. What does matter is which singularities the loop encloses and how many times it winds around each: those are topological data. This is why in practice one picks whatever contour is convenient — a large semicircle to close a real-axis integral, a rectangle for Matsubara sums, a keyhole around a branch cut — confident that only the enclosed poles and their winding numbers can change the answer.