Cauchy Integral Formula

Why the Cauchy integral formula matters

• Entire function theory. Liouville's theorem, Picard's theorems, Hadamard factorization — every classical result on entire functions runs through Cauchy's formula. The order, type, and growth of an entire function are extracted directly from boundary integrals.
• Residue theorem. The residue theorem is the punctured generalization: replace 1/(z−a) with f(z), expand f as a Laurent series around each pole, and the formula collapses to 2πi times the sum of residues. The two are the same theorem read in different directions.
• Fundamental theorem of algebra. The two-line FTA proof — Liouville applied to 1/p(z) — is the canonical advertisement for complex methods. No real-variable proof of comparable shortness exists.
• Contour integration. Real integrals like ∫₀^∞ sin(x)/x dx, ∫₀^∞ x^(a−1)/(1+x) dx, and the Beta function all evaluate via Cauchy's formula on a cleverly chosen contour. The formula is the backbone of the residue method.
• Numerical analysis. The trapezoidal rule on a circle has exponential convergence for periodic analytic functions — a direct consequence of the Cauchy formula's power-series expansion. Spectral methods and Trefethen's "Approximation Theory and Approximation Practice" build on this.
• PDE theory. Solutions to elliptic PDEs in the disc are recovered from boundary data via the Poisson integral, the harmonic-function analogue of Cauchy's formula. Dirichlet's problem, Schwarz reflection, and conformal welding all descend from this lineage.
• Random matrix theory. Resolvents (zI − A)⁻¹ and contour integrals of their traces give spectral information; eigenvalue distributions of large random matrices are extracted via Cauchy-style boundary integrals.

Common misconceptions

• "Needs disc-shaped contour." Any simple closed contour works as long as it bounds a region in the simply connected domain. Triangles, squares, ellipses, smooth Jordan curves — all valid.
• "Real analysis has the same property." No. Real C^∞ functions are not determined by boundary values; bump functions are smooth and zero off a compact set. The Cauchy formula is a uniquely complex phenomenon driven by the Cauchy-Riemann equations' rigidity.
• "f must be smooth on the closure." The formula needs only that f is holomorphic in an open set containing the closed region bounded by γ. Boundary regularity is not required beyond holomorphicity in a neighborhood.
• "The formula gives f everywhere." Only inside γ. For points a outside the region bounded by γ, the integrand z ↦ f(z)/(z−a) is holomorphic on the inside, so the integral is zero by Cauchy-Goursat. Direction matters: outside means zero, inside means f(a).
• "Differentiating under the integral is hand-wavy." It is rigorously justified by uniform convergence of difference quotients on compact subsets bounded away from γ. The derivative formula f^(n)(a) = (n!/2πi) ∮ f(z)/(z−a)^(n+1) dz follows by induction.
• "Liouville bounds growth, not values." Liouville is sharp. e^z is entire and unbounded; sin(z) is entire and unbounded on ℂ (though bounded on ℝ); 1/(1+z²) is not entire because of poles at ±i. Bounded entire functions really are constants — the conclusion is global.

Statement and immediate consequences

Let D ⊂ ℂ be a simply connected open set, f: D → ℂ holomorphic, and γ a positively oriented simple closed contour in D enclosing a point a ∈ D. Then

f(a) = (1 / 2πi) · ∮_γ f(z) / (z − a) dz.

Differentiating both sides n times with respect to a (and rigorously moving the derivative inside the integral) gives the derivative formulas:

f^(n)(a) = (n! / 2πi) · ∮_γ f(z) / (z − a)^(n+1) dz   for every n ≥ 0.

From these two lines:

• Holomorphic ⇒ infinitely differentiable. The right-hand side defines all derivatives in terms of the integral, so every holomorphic f has derivatives of all orders.
• Holomorphic ⇒ analytic. Expanding 1/(z−a) as a power series in a around a base point a₀, swapping sum and integral, gives a Taylor series for f around a₀ with radius of convergence at least dist(a₀, ∂D).
• Cauchy estimates. If |f| ≤ M on a circle of radius R around a, then |f^(n)(a)| ≤ n! M / Rⁿ. This bound, which has no real-analysis analogue, is the workhorse for everything that follows.

Liouville's theorem and the FTA

Liouville's theorem says every bounded entire function is constant. The proof is one Cauchy estimate.

Suppose f is entire and |f(z)| ≤ M for all z ∈ ℂ.
For any point a and any radius R > 0, take γ_R = circle of radius R around a.
By the Cauchy estimate at order 1:
        |f′(a)| ≤ M / R.
Let R → ∞: |f′(a)| ≤ 0, so f′(a) = 0.
This holds for every a, so f′ ≡ 0 on ℂ, hence f is constant.

The Fundamental Theorem of Algebra drops out in two lines. Suppose p(z) is a polynomial of degree n ≥ 1 with no zeros. Then 1/p(z) is entire. As |z| → ∞, |p(z)| → ∞ (the leading term dominates), so 1/|p(z)| → 0. In particular 1/p is bounded on a large disc, and tends to zero outside, so 1/p is a bounded entire function. By Liouville, 1/p is constant, hence p is constant — contradicting deg p ≥ 1. Therefore every non-constant polynomial has at least one zero in ℂ.

Iterating (factor out (z − z₀), repeat) shows p has exactly n zeros counted with multiplicity. The complex numbers are algebraically closed.

The maximum modulus principle

Let f be holomorphic and non-constant on a connected open set U. Then |f| has no local maximum in U; equivalently, on any compact set K ⊂ U the maximum of |f| is attained on ∂K.

Proof sketch: take a ∈ U with |f(a)| ≥ |f(z)| for all z in a small neighborhood. By Cauchy's integral formula on a small circle of radius r around a,

|f(a)| = |(1/2π) ∫₀^{2π} f(a + r e^{iθ}) dθ|
       ≤ (1/2π) ∫₀^{2π} |f(a + r e^{iθ})| dθ
       ≤ |f(a)|.

Equality forces |f| to be constant on the circle, then on the disc (varying r), then on U (identity theorem). So a non-constant f cannot attain a local maximum of |f| in U.

Practical consequences: numerical solvers for elliptic PDEs converge with error bounds determined entirely by boundary data; conformal mappings preserve maximum modulus; Schwarz lemma (a refinement on the disc) gives the rigid form |f(z)| ≤ |z| for f: 𝔻 → 𝔻 with f(0) = 0.

Proof from Cauchy-Goursat

Cauchy-Goursat: if f is holomorphic on a simply connected D and γ is closed in D, then ∮_γ f dz = 0. Take this as given (its standard proof uses subdivision into triangles plus a careful diameter estimate).

To prove the Cauchy integral formula, fix a inside the region bounded by γ. The function f(z)/(z−a) is holomorphic on D \ {a}. Cut a small circle C_ε of radius ε around a, oriented positively. The region between γ and C_ε is bounded by γ and −C_ε (C_ε reversed), and f(z)/(z−a) is holomorphic there. By Cauchy-Goursat applied to that annular region,

∮_γ f(z)/(z−a) dz = ∮_{C_ε} f(z)/(z−a) dz.

On C_ε, parametrize z = a + ε e^{iθ}, dz = i ε e^{iθ} dθ:

∮_{C_ε} f(z)/(z−a) dz = ∫₀^{2π} f(a + ε e^{iθ}) · (i ε e^{iθ}) / (ε e^{iθ}) dθ
                       = i · ∫₀^{2π} f(a + ε e^{iθ}) dθ.

Let ε → 0. By continuity of f, f(a + ε e^{iθ}) → f(a) uniformly in θ. The integral converges to i · 2π · f(a). Therefore

∮_γ f(z)/(z−a) dz = 2πi · f(a),

which rearranges to the formula. The proof is two lines plus a parametrization.

Extensions and refinements

• Cauchy estimates and growth orders. The order of an entire function f is ρ = limsup log log M(r) / log r, where M(r) = max |f| on |z| = r. Cauchy estimates relate Taylor coefficients to M(r) and underpin the Hadamard factorization theorem (entire functions of finite order are products of canonical exponentials and primary factors over their zeros).
• Schwarz reflection principle. If f is holomorphic on the upper half disc and continuous on the real diameter where it takes real values, then f extends holomorphically to the full disc by f(z̄) = f(z) overline. The proof uses Cauchy's formula and the symmetry of contours under reflection.
• Cauchy's formula on the boundary (Plemelj-Sokhotski). When a sits on γ rather than inside, the integral takes a principal-value form and gives (1/2)[f(a) + (1/iπ) PV ∮ f(z)/(z−a) dz]. This is the foundation of the Hilbert transform and singular integral operators in Riemann-Hilbert problems.
• Cauchy formula on Riemann surfaces. Replace γ by a closed curve in a Riemann surface, and the integrand by a meromorphic differential; the formula generalizes to give residues of differentials, leading to the Riemann-Roch theorem.
• Bochner-Martinelli (multivariate analogue). In ℂⁿ, the Bochner-Martinelli kernel generalizes 1/(z−a); for holomorphic f on a domain in ℂⁿ, f(a) is recovered as a boundary integral of a more elaborate kernel. Foundation of several-complex-variable theory.
• Functional calculus on operators. For a bounded operator A on a Banach space and a holomorphic f on a contour enclosing the spectrum of A, define f(A) = (1/2πi) ∮ f(z)(zI − A)⁻¹ dz. The Cauchy formula generalizes to operator-valued resolvents and is the backbone of Dunford-Riesz functional calculus and spectral theory.

Where the formula shows up

• Quantum field theory — propagators. Free-field propagators G(x − y) are evaluated as contour integrals over momentum; the iε prescription chooses which poles get enclosed, and Cauchy's formula extracts the on-shell contribution. Feynman, Wick, and Mandelstam representations are all Cauchy integrals dressed differently.
• Control theory — stability of linear systems. The Nyquist stability criterion encloses poles of the closed-loop transfer function; the argument principle (a Cauchy formula consequence applied to f′/f) counts encirclements equal to unstable poles. Engineers draw Nyquist plots without knowing they are reading off Cauchy integrals.
• Signal processing — analyticity of Z-transforms. Causal discrete-time signals have Z-transforms holomorphic outside a disc; inverse transforms are computed by Cauchy contour integrals, and FIR/IIR filter design uses pole-zero placement informed by maximum modulus arguments.
• Numerical analysis — trapezoidal rule on circles. For periodic analytic f, the trapezoidal rule on a circle has error decaying exponentially in the number of nodes (rather than polynomially as for generic smooth f). The proof reads off Cauchy's formula. Trefethen's spectral methods exploit this for chebfun and similar libraries.
• Asymptotic analysis — saddle-point method. The method of stationary phase and the saddle-point method approximate ∮ f(z) e^{N g(z)} dz for large N by deforming γ through a saddle of g. Cauchy's formula and Cauchy-Goursat justify the deformation; the asymptotics extract leading terms via local Gaussian integrals.
• Combinatorics — generating function coefficient extraction. If F(z) = Σ aₙ zⁿ, then aₙ = (1/2πi) ∮ F(z)/z^(n+1) dz on any circle inside the radius of convergence. Asymptotic enumeration (Flajolet-Sedgewick) reads off coefficient growth from singularities of F via this exact identity.

Worked examples

Example 1: Compute ∮_γ e^z/(z − 2) dz where γ is the circle |z| = 3. The function e^z is entire; the singularity at z = 2 lies inside γ; by Cauchy's formula the integral equals 2πi · e² ≈ 46.42 i.

Example 2: Compute ∮_γ cos(z)/(z² + 1) dz where γ is the circle |z| = 2. Factor 1/(z² + 1) = 1/[(z − i)(z + i)]; both poles ±i lie inside γ. Apply the formula at each pole. Residue at i: cos(i)/(2i) = cosh(1)/(2i). Residue at −i: cos(−i)/(−2i) = cosh(1)/(−2i). Sum of residues: cosh(1)/(2i) − cosh(1)/(2i) = 0? No — careful, residues add: cosh(1)/(2i) + cosh(1)/(−2i) wait — let me redo. Res(cos(z)/(z²+1), i) = cos(i)/(2i), Res at −i = cos(−i)/(−2i) = cos(i)/(−2i). Sum = cos(i)·(1/(2i) − 1/(2i)) = 0. So ∮ = 0. (The contour encloses both conjugate poles whose residues cancel.)

Example 3: Derive ζ(2) = π²/6 by the Cauchy formula trick. Define F(z) = π cot(πz)/z². Poles of F: simple at every nonzero integer (from cot), order-3 at 0 (from cot's pole at 0 and the z² factor). Sum of residues over an expanding square contour vanishes (because |cot(πz)| is bounded on the squares away from integers). So 0 = Res(F, 0) + 2 · Σₙ₌₁^∞ 1/n². The residue at 0 is the (z²)·(1/z) coefficient of π cot(πz), which is −π²/3 (Laurent expansion). Hence Σ 1/n² = π²/6.