Real differentiability is plentiful but cheap. Every function $f: \mathbb{R}^2 \to \mathbb{R}^2$ that happens to have continuous first-order partial derivatives is called $C^1$, and there are vastly many such functions — ones that twist, compress, and shear in wildly incoherent ways. Yet among all smooth maps from the plane to itself, a very special subclass emerges when we impose one additional constraint: that the derivative at every point be not merely linear, but *complex*-linear. This single requirement — that the map be compatible with multiplication by $i$ — turns out to enforce an astonishing cascade of properties. Functions in this subclass are infinitely differentiable, representable by convergent power series, completely determined by their values on any curve, and governed by one of the most powerful integral formulas in all of mathematics. These are the holomorphic functions, and they are the central objects of complex analysis.
The contrast with real analysis is stark. A real-analytic function on an interval can be prescribed almost freely on a subinterval and extended in multiple ways beyond it. A holomorphic function on a connected open set, by contrast, is completely determined by its values on any open subdisk — or even on any convergent sequence of points. This "rigidity" is not a limitation; it is what makes complex analysis so effective. Questions that seem intractable in real analysis often reduce cleanly to contour integrals, residue calculations, and conformal mappings once holomorphicity is available.
[example: A Differentiable Function That Is Not Holomorphic]
Consider the function
\begin{align*}
f: \mathbb{C} &\to \mathbb{C} \\
z &\mapsto \bar{z}.
\end{align*}
Write $z = x + iy$ and $\bar{z} = x - iy$. As a map $\mathbb{R}^2 \to \mathbb{R}^2$, this is the reflection $(x, y) \mapsto (x, -y)$, which is smooth ($C^\infty$) with Jacobian matrix
\begin{align*}
Jf &= \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}.
\end{align*}
In particular the real partial derivatives all exist and are constant. Nevertheless, $f(z) = \bar{z}$ is not holomorphic at any point. To see why, examine what "complex differentiability" requires: the limit
\begin{align*}
\lim_{h \to 0} \frac{f(z + h) - f(z)}{h} &= \lim_{h \to 0} \frac{\overline{z + h} - \bar{z}}{h} = \lim_{h \to 0} \frac{\bar{h}}{h}
\end{align*}
must exist as $h \to 0$ through all complex values. Taking $h = t \in \mathbb{R} \setminus \{0\}$ gives $\bar{h}/h = t/t = 1$. Taking $h = it$ with $t \in \mathbb{R} \setminus \{0\}$ gives $\bar{h}/h = -it/it = -1$. Since the limit along the real axis equals $1$ and the limit along the imaginary axis equals $-1$, the two-dimensional limit does not exist. The function $\bar{z}$ therefore has no complex derivative anywhere, despite being a perfectly smooth real map.
This example reveals what complex differentiability is really demanding: not just the existence of partial derivatives, but a *specific algebraic relationship* between them.
[/example]
## Definition
The example above shows that the naive real-differentiability condition is too permissive. What we need is for the linear approximation to $f$ at each point to commute with multiplication by $i$. In terms of the real and imaginary parts, this translates into a pair of partial differential equations.
[definition: Holomorphic Function]
Let $\Omega \subset \mathbb{C}$ be an open set, and let $f: \Omega \to \mathbb{C}$ be a function. Write $f(x + iy) = u(x, y) + iv(x, y)$ where $u, v: \Omega \to \mathbb{R}$ are real-valued. We say $f$ is **holomorphic on $\Omega$** if the partial derivatives $\partial_x u$, $\partial_y u$, $\partial_x v$, $\partial_y v$ exist and are continuous on $\Omega$, and satisfy the **Cauchy–Riemann equations**:
\begin{align*}
\partial_x u &= \partial_y v, \\
\partial_y u &= -\partial_x v.
\end{align*}
[/definition]
The space of holomorphic functions on $\Omega$ is denoted $\mathcal{O}(\Omega)$.
The Cauchy–Riemann equations are the analytic expression of complex linearity of the derivative. To see this, recall that the total derivative $Df_z: \mathbb{R}^2 \to \mathbb{R}^2$ at a point $z = x + iy$ is represented by the Jacobian matrix
\begin{align*}
Jf_z &= \begin{pmatrix} \partial_x u & \partial_y u \\ \partial_x v & \partial_y v \end{pmatrix}.
\end{align*}
Multiplication by a complex number $\alpha = a + ib$ corresponds, via the identification $\mathbb{C} \cong \mathbb{R}^2$, to left-multiplication by the matrix $\begin{pmatrix} a & -b \\ b & a \end{pmatrix}$. For $Jf_z$ to represent complex multiplication, it must have exactly this form: the diagonal entries equal and the off-diagonal entries negatives of each other. This is precisely the Cauchy–Riemann condition.
[remark: Wirtinger Derivatives]
The Cauchy–Riemann equations have a compact reformulation using the Wirtinger derivatives:
\begin{align*}
\partial_z &= \frac{1}{2}(\partial_x - i\partial_y), \qquad \partial_{\bar{z}} = \frac{1}{2}(\partial_x + i\partial_y).
\end{align*}
A function $f \in C^1(\Omega)$ is holomorphic if and only if $\partial_{\bar{z}} f = 0$ on $\Omega$. The complex derivative then equals $f'(z) = \partial_z f$. This formalism makes it immediate that holomorphic functions are exactly those that "do not depend on $\bar{z}$."
[/remark]
When $f$ is holomorphic, the complex derivative $f'(z) = \partial_x u + i\partial_x v$ exists at every point of $\Omega$ and equals the limit
\begin{align*}
f'(z) = \lim_{h \to 0} \frac{f(z + h) - f(z)}{h}
\end{align*}
taken through all complex $h \to 0$. The two definitions — Cauchy–Riemann plus $C^1$, or existence of the complex derivative — are equivalent for $C^1$ functions. The deeper theorem, which we will encounter below, is that complex differentiability at every point of an open set is already enough to force $f \in C^\infty(\Omega)$.
[example: Basic Holomorphic Functions]
The most fundamental examples come from the basic operations of complex arithmetic.
**Polynomials.** Any polynomial $p(z) = a_n z^n + \cdots + a_1 z + a_0$ with $a_k \in \mathbb{C}$ is holomorphic on all of $\mathbb{C}$. To verify for $z^n$: writing $z = x + iy$, the function $u + iv = (x + iy)^n$ satisfies the Cauchy–Riemann equations at every point by the complex chain rule. Alternatively, the limit $\lim_{h \to 0} (z+h)^n/h - z^n/h = nz^{n-1}$ exists for all $z$.
**The exponential.** Define $e^z := e^x(\cos y + i \sin y)$ for $z = x + iy$. Then $u = e^x \cos y$ and $v = e^x \sin y$. Computing: $\partial_x u = e^x \cos y = \partial_y v$ and $\partial_y u = -e^x \sin y = -\partial_x v$. The Cauchy–Riemann equations hold everywhere, so $e^z \in \mathcal{O}(\mathbb{C})$.
**The logarithm.** The function $\log z$ cannot be defined as a holomorphic function on all of $\mathbb{C} \setminus \{0\}$, because any continuous logarithm on a loop around the origin would have to increment by $2\pi i$ — a topological obstruction. On the slit plane $\Omega = \mathbb{C} \setminus (-\infty, 0]$, define the principal branch $\operatorname{Log}(z) = \ln|z| + i\operatorname{Arg}(z)$ where $\operatorname{Arg}(z) \in (-\pi, \pi)$. One verifies that $\operatorname{Log} \in \mathcal{O}(\Omega)$ with derivative $1/z$.
[/example]
## The Cauchy Integral Theorem and Its Consequences
The single most powerful result in complex analysis is that holomorphic functions satisfy a remarkable global rigidity property: their values at every point of a domain are completely encoded in their boundary values, and are recoverable by a contour integral. To reach this, we first need the Cauchy integral theorem, which says that contour integrals of holomorphic functions around closed curves vanish.
Before stating the theorem precisely, it is worth pausing to understand what makes this surprising. For a real function $f: [a,b] \to \mathbb{R}$, the integral $\int_a^b f(x)\, dx$ depends on the path (there is only one path, the interval). For a smooth real function on $\mathbb{R}^2$, the line integral $\int_\gamma f\, dx + g\, dy$ generally depends on the path $\gamma$. For holomorphic functions, path independence of contour integrals is automatic — not because the domain is special, but because the Cauchy–Riemann equations force a miraculous cancellation.
[quotetheorem:3352]
The proof proceeds by writing $f\, dz = (u + iv)(dx + i\, dy) = (u\, dx - v\, dy) + i(v\, dx + u\, dy)$ and applying Green's theorem to each real line integral. The Cauchy–Riemann equations ensure that the integrand of each Green's theorem term vanishes. Simple connectivity is used to ensure that the interior of $\gamma$ lies entirely within $\Omega$.
[illustration:simply-connected-vs-multiply-connected]
From the Cauchy integral theorem, the Cauchy integral formula follows by a short but elegant argument: one isolates the singularity $1/(z - z_0)$ and shows that the entire contribution comes from a tiny circle around $z_0$.
[quotetheorem:345]
This formula has consequences that seem almost magical from the standpoint of real analysis. Because the right-hand side is an integral of $f$ over a fixed circle, one can differentiate under the integral sign with respect to $z$ as many times as desired. Each differentiation produces a new formula for $f^{(n)}(z)$, showing that $f$ is infinitely differentiable.
[quotetheorem:3353]
This is a complete contrast to real analysis, where a function can be $C^1$ but not $C^2$. In the complex world, being $C^1$ in the complex sense forces $C^\infty$ in the real sense. The derivative formula above also gives immediate estimates: if $|f(w)| \le M$ on the circle $|w - z_0| = r$, then
\begin{align*}
|f^{(n)}(z_0)| \le \frac{n! M}{r^n}.
\end{align*}
These are the **Cauchy estimates**, and they are the engine behind several of the deepest theorems in complex analysis.
## Power Series and Analytic Functions
The Cauchy integral formula not only shows that $f$ is infinitely differentiable; it shows that $f$ is represented locally by a convergent power series. This is the content of the theorem that holomorphic equals analytic.
Before stating this, it is worth understanding why the power series representation is nontrivial. In real analysis, $C^\infty$ functions need not be real-analytic: the function
\begin{align*}
\phi(x) &:= \begin{cases} e^{-1/x^2} & x > 0 \\ 0 & x \le 0 \end{cases}
\end{align*}
is $C^\infty$ on $\mathbb{R}$ but its Taylor series at $x = 0$ converges to $0$ everywhere, not to $\phi$. To see why, note that $\phi^{(n)}(0) = 0$ for every $n \ge 0$: for $x > 0$ this follows by induction, since any derivative of $e^{-1/x^2}$ has the form $P(1/x)e^{-1/x^2}$ for some polynomial $P$, and $e^{-1/x^2}$ decays faster than any power of $x$ as $x \to 0^+$ (indeed $t^k e^{-t} \to 0$ as $t \to \infty$ for any $k$, so $x^{-k}e^{-1/x^2} = (1/x)^k e^{-1/x^2} \to 0$ as $x \to 0^+$). Since all derivatives at $0$ vanish, the Taylor series $\sum_{n=0}^\infty (\phi^{(n)}(0)/n!) x^n = 0$ is identically zero, which does not equal $\phi(x)$ for $x > 0$. In the complex world, this failure is impossible.
[quotetheorem:3354]
The proof extracts the geometric series $1/(w - z) = \sum_{n=0}^\infty (z - z_0)^n / (w - z_0)^{n+1}$ (valid for $|z - z_0| < |w - z_0|$) from the Cauchy integral formula and integrates term by term, justifying the interchange by uniform convergence on the circle of integration.
Conversely, any function defined by a convergent power series is holomorphic wherever the series converges, since power series are termwise differentiable. The two classes — holomorphic functions and convergent power series — are therefore the same. This equivalence is sometimes taken as the definition of holomorphic (under the name "analytic"), but the Cauchy–Riemann formulation is more natural for relating complex analysis to real analysis and PDE theory.
[example: Radius of Convergence as Distance to the Nearest Singularity]
Consider $f(z) = 1/(1 - z)$, holomorphic on $\mathbb{C} \setminus \{1\}$. Expanding around $z_0 = 0$ gives the geometric series
\begin{align*}
\frac{1}{1 - z} &= \sum_{n=0}^\infty z^n,
\end{align*}
which converges for $|z| < 1$. The radius of convergence is $1$, which equals $|z_0 - 1| = |0 - 1|$, the distance from the center $z_0 = 0$ to the nearest singularity at $z = 1$.
Now expand around $z_0 = i$. Writing $1 - z = 1 - i - (z - i)$, we have
\begin{align*}
\frac{1}{1 - z} &= \frac{1}{(1 - i)(1 - (z-i)/(1-i))} = \frac{1}{1-i} \sum_{n=0}^\infty \left(\frac{z-i}{1-i}\right)^n,
\end{align*}
which converges for $|z - i| < |1 - i| = \sqrt{2}$. Indeed $|z_0 - 1| = |i - 1| = \sqrt{2}$, confirming the pattern. In general, for a holomorphic function $f$ defined on all of $\mathbb{C}$ except for isolated singularities, the radius of convergence of the Taylor series around $z_0$ is exactly the distance from $z_0$ to the nearest singularity. This has no analogue in real analysis: the function $1/(1 + x^2)$ is $C^\infty$ on all of $\mathbb{R}$, but its Taylor series at $x = 0$ has radius of convergence only $1$, because the complex singularities at $z = \pm i$ are distance $1$ from the real axis.
[/example]
## Isolated Singularities and Laurent Series
Holomorphic functions behave beautifully on their domains of definition, but the most interesting questions in complex analysis often involve what happens near points where holomorphicity fails. [definition: Isolated Singularity]
Let $f: \Omega \to \mathbb{C}$ be holomorphic on an open set $\Omega \subset \mathbb{C}$. A point $z_0 \in \mathbb{C}$ is an **isolated singularity** of $f$ if there exists $r > 0$ such that $f$ is holomorphic on the punctured disk $B(z_0, r) \setminus \{z_0\}$ but $f$ is not holomorphic at $z_0$ itself.
[/definition]
Understanding what $f$ does near $z_0$ requires a finer expansion than a power series.
Near an isolated singularity, the power series must be supplemented by negative powers of $(z - z_0)$. The reason is simple: a function that blows up at $z_0$ cannot be expanded in non-negative powers of $(z - z_0)$, because such a series would converge to a bounded quantity near $z_0$. What we need is a two-sided expansion — one that allows arbitrarily negative powers to capture the singular behavior, while the non-negative powers handle the regular part. This is precisely what the Laurent series provides, and it is the natural generalization of the Taylor series to functions with isolated singularities.
[definition: Laurent Series]
Let $z_0 \in \mathbb{C}$ and $0 \le r_1 < r_2 \le \infty$. A **Laurent series** centered at $z_0$ on the annulus $A = \{z \in \mathbb{C} : r_1 < |z - z_0| < r_2\}$ is a series of the form
\begin{align*}
\sum_{n=-\infty}^{\infty} c_n (z - z_0)^n,
\end{align*}
where $c_n \in \mathbb{C}$ for all $n \in \mathbb{Z}$. The **principal part** of the series is $\sum_{n=-\infty}^{-1} c_n (z - z_0)^n$.
[/definition]
Every holomorphic function on an annulus has a Laurent expansion, given by a formula analogous to the Taylor coefficient formula. The behavior of the principal part at $z_0$ classifies the singularity, and the three possibilities — removable singularity, pole, and essential singularity — are genuinely different phenomena that require separate names and separate techniques.
The mildest case is the removable singularity. Here the singularity is an artifact of how $f$ was defined, not a genuine failure of holomorphicity: the function simply needs to be assigned the right value at $z_0$ to become holomorphic there. The Laurent expansion detects this because the principal part — the negative-power terms — is what obstructs the existence of a holomorphic extension. If there is no principal part at all, there is nothing standing in the way of extension.
[definition: Removable Singularity]
An isolated singularity $z_0$ of $f$ is **removable** if there exists a holomorphic function $g: B(z_0, r) \to \mathbb{C}$ such that $g(z) = f(z)$ for all $z \in B(z_0, r) \setminus \{z_0\}$.
[/definition]
The principal part of the Laurent expansion characterizes removability: $z_0$ is a removable singularity if and only if the principal part vanishes, i.e., $c_n = 0$ for all $n \le -1$. In other words, the Laurent expansion reduces to an ordinary power series, which converges at $z_0$ and provides the missing value $g(z_0) = c_0$.
When the principal part is nonempty but finite — finitely many negative-power terms — we have a qualitatively different situation: the function genuinely blows up at $z_0$, but in a controlled, polynomial way. These are the poles, and they are the singularities that are easiest to work with in practice. A pole of order $m$ means $f(z)$ grows like $1/(z - z_0)^m$ near $z_0$, and the function $(z - z_0)^m f(z)$ extends to a nonzero holomorphic function at $z_0$.
[definition: Pole]
An isolated singularity $z_0$ of $f$ is a **pole of order $m \in \mathbb{N}$** if the Laurent expansion of $f$ around $z_0$ has the form
\begin{align*}
f(z) &= \frac{c_{-m}}{(z - z_0)^m} + \frac{c_{-m+1}}{(z - z_0)^{m-1}} + \cdots + \frac{c_{-1}}{z - z_0} + c_0 + c_1(z - z_0) + \cdots,
\end{align*}
where $c_{-m} \ne 0$. A pole of order $1$ is called a **simple pole**.
[/definition]
The third possibility is the most dramatic. When the principal part has infinitely many nonzero terms, no finite polynomial in $1/(z - z_0)$ can capture the singular behavior. The function oscillates wildly near $z_0$, taking values spread throughout $\mathbb{C}$ in every punctured neighborhood. This is the essential singularity, and it is in every sense the worst-behaved case.
[definition: Essential Singularity]
An isolated singularity $z_0$ of $f$ is an **essential singularity** if the principal part of the Laurent expansion around $z_0$ has infinitely many nonzero terms, i.e., $c_n \ne 0$ for infinitely many $n \le -1$.
[/definition]
These three types of singularity exhibit radically different behaviors. A removable singularity is, in a precise sense, not really a singularity at all — $f$ simply needs to be assigned the right value at $z_0$. A pole causes $|f(z)| \to \infty$ as $z \to z_0$. An essential singularity is far more dramatic: near such a point, the function does not merely blow up or tend to a limit — it visits every complex value, or comes arbitrarily close to doing so, in every punctured neighborhood.
[quotetheorem:3355]
The Casorati–Weierstrass theorem shows that essential singularities produce chaotic behavior — but what about the opposite extreme? If $f$ is well-behaved near $z_0$, one expects the singularity to be removable. The following theorem makes this precise: boundedness alone is enough to guarantee that $f$ extends holomorphically across $z_0$.
[quotetheorem:3356]
The key observation in the proof is that if $f$ is bounded near $z_0$, then $g(z) := (z - z_0)^2 f(z)$ extends holomorphically to $z_0$ with $g(z_0) = g'(z_0) = 0$. The function $h(z) = g(z)/(z-z_0)^2$ then extends to $z_0$ by L'Hôpital-type reasoning.
The Laurent expansion encodes everything about the behavior of $f$ near an isolated singularity, but one coefficient stands out above all others: the coefficient of $(z - z_0)^{-1}$. The reason for its special status is integration. Every other term $c_n(z - z_0)^n$ with $n \ne -1$ has an antiderivative $c_n(z - z_0)^{n+1}/(n+1)$ that is single-valued on the punctured disk, so it contributes zero to any closed contour integral around $z_0$. The term $c_{-1}/(z - z_0)$ alone has no single-valued antiderivative — its antiderivative would be $c_{-1} \log(z - z_0)$, which is multi-valued. When integrated around a small circle, it contributes $2\pi i c_{-1}$. This single number, the residue, is therefore the key to evaluating contour integrals.
[definition: Residue]
Let $z_0$ be an isolated singularity of $f$ with Laurent expansion $\sum_{n=-\infty}^\infty c_n(z-z_0)^n$ on a punctured disk. The **residue** of $f$ at $z_0$ is the coefficient
\begin{align*}
\operatorname{Res}(f, z_0) &:= c_{-1}.
\end{align*}
[/definition]
The residue is singled out among all the Laurent coefficients because it is the only one that survives integration around a small circle: for any small enough circle $\gamma$ around $z_0$, the term $c_n(z-z_0)^n$ integrates to $0$ for $n \ne -1$ (by direct computation, since the antiderivative $(z-z_0)^{n+1}/(n+1)$ is single-valued), while the term $c_{-1}/(z-z_0)$ integrates to $2\pi i c_{-1}$. This makes the residue the key invariant for computing contour integrals.
[quotetheorem:352]
[illustration:semicircular-contour]
[example: Computing a Real Integral via Residues]
We compute
\begin{align*}
I &= \int_{-\infty}^{\infty} \frac{1}{1 + x^2}\, dx
\end{align*}
using the residue theorem. Consider $f(z) = 1/(1 + z^2)$. Factor: $1 + z^2 = (z - i)(z + i)$, so $f$ has simple poles at $z = i$ and $z = -i$. Let $\gamma_R$ be the positively oriented contour consisting of the segment $[-R, R]$ on the real axis and the upper semicircle $\Gamma_R = \{Re^{i\theta} : 0 \le \theta \le \pi\}$, for $R > 1$.
The contour $\gamma_R$ encloses only the pole at $z = i$. The residue there is
\begin{align*}
\operatorname{Res}(f, i) &= \lim_{z \to i} (z - i) \cdot \frac{1}{(z-i)(z+i)} = \frac{1}{2i}.
\end{align*}
By the residue theorem:
\begin{align*}
\oint_{\gamma_R} f(z)\, dz &= 2\pi i \cdot \frac{1}{2i} = \pi.
\end{align*}
On the semicircle $\Gamma_R$, parametrize $z = Re^{i\theta}$, so $|dz| = R\, d\theta$ and $|f(z)| \le 1/(R^2 - 1)$ for $R > 1$. Therefore
\begin{align*}
\left|\int_{\Gamma_R} f(z)\, dz\right| &\le \frac{\pi R}{R^2 - 1} \to 0 \quad \text{as } R \to \infty.
\end{align*}
Splitting the contour integral:
\begin{align*}
\pi &= \oint_{\gamma_R} f(z)\, dz = \int_{-R}^{R} \frac{1}{1 + x^2}\, dx + \int_{\Gamma_R} f(z)\, dz.
\end{align*}
Sending $R \to \infty$, the semicircle integral vanishes, giving $I = \pi$.
[/example]
## Conformal Mappings and the Open Mapping Theorem
A holomorphic function with everywhere nonzero derivative does something remarkable to the geometry of the plane: it preserves angles. If two smooth curves meet at a point $z_0$ at angle $\alpha$, their images under a holomorphic $f$ with $f'(z_0) \ne 0$ also meet at angle $\alpha$ — and in the same orientation. Such a map is called conformal, and the study of conformal mappings is one of the most geometrically rich parts of complex analysis.
The reason angle preservation follows from the nonvanishing derivative is that near $z_0$, $f$ looks like the linear map $w \mapsto f'(z_0)(w - z_0)$, which is multiplication by the complex number $f'(z_0)$. Multiplication by a nonzero complex number rotates and scales, but rotation preserves angles. The nonvanishing of $f'(z_0)$ is what ensures the map is not collapsing curves onto a point, which would destroy angle information.
Angle-preservation captures the local behavior at a point, but we often want more: a bijective holomorphic map between two domains that is holomorphic in both directions. Such a map sets up a perfect complex-analytic equivalence between the domains — every holomorphic object on one transfers to the other. The question of which domains are conformally equivalent turns out to be a deep classification problem, answered in the simply connected case by the Riemann mapping theorem. These considerations motivate isolating two closely related but distinct notions in a single definition.
[definition: Conformal Map]
Let $\Omega \subset \mathbb{C}$ be open. A holomorphic function $f \in \mathcal{O}(\Omega)$ is **conformal at $z_0 \in \Omega$** if $f'(z_0) \ne 0$. It is **conformal on $\Omega$** if $f'(z) \ne 0$ for all $z \in \Omega$. A bijective conformal map $f: \Omega \to \Omega'$ is called a **biholomorphism** or **conformal equivalence**; $\Omega$ and $\Omega'$ are then said to be conformally equivalent.
[/definition]
The canonical example is the Möbius transformation $f(z) = (az + b)/(cz + d)$ with $ad - bc \ne 0$, which is conformal and bijective on the Riemann sphere $\mathbb{C} \cup \{\infty\}$. More generally, every conformal automorphism of the upper half-plane or the unit disk is a Möbius transformation.
One might hope that a conformal map sends open sets to open sets — after all, it is locally a rotation and scaling, which certainly preserves openness. The open mapping theorem confirms this, and in fact proves something stronger.
[quotetheorem:712]
The proof proceeds by showing that for any $z_0 \in U$ and $w_0 = f(z_0)$, the function $f(z) - w_0$ has an isolated zero at $z_0$, and then counting zeros of $f(z) - w$ for $w$ near $w_0$ via the argument principle.
The argument principle itself is a beautiful application of residues: it counts the zeros and poles of a meromorphic function inside a contour by computing a single integral.
[quotetheorem:356]
The key observation in the proof is that $f'/f$ has a simple pole at each zero of $f$ with residue equal to the order, and a simple pole at each pole of $f$ with residue equal to the negative of the order. The residue theorem then converts the contour integral into the zero–pole count.
A striking consequence of the argument principle is Rouché's theorem, which allows the zero count of a holomorphic function to be estimated by comparison with a simpler function.
[quotetheorem:357]
Rouché's theorem is used to locate zeros: by choosing $f$ to be the dominant term of a function, one can count the zeros of a complicated function by counting the zeros of a simple one. For instance, the polynomial $p(z) = z^5 + 3z + 1$ has all five zeros inside the disk $|z| < 2$, because on $|z| = 2$ we have $|z^5| = 32 > |3z + 1| \le 7$, so $z^5$ and $z^5 + 3z + 1$ have the same zero count inside $|z| = 2$.
Conformal mappings connect complex analysis to a wide range of applications. Two are worth developing in some detail.
**Fluid dynamics.** Consider a two-dimensional, incompressible, irrotational fluid flow. Incompressibility requires $\operatorname{div}v = \partial_x v_1 + \partial_y v_2 = 0$, and irrotationality requires $\operatorname{curl}v = \partial_x v_2 - \partial_y v_1 = 0$. These are precisely the Cauchy–Riemann equations for the complex function $F'(z) = v_1 - iv_2$: the real part $v_1$ and the imaginary part $-v_2$ satisfy $\partial_x v_1 = \partial_y(-v_2)$ and $\partial_y v_1 = -\partial_x(-v_2)$. Thus the complex velocity potential $F(z)$ is holomorphic, and the flow is completely encoded in $F$. A conformal map $w = \phi(z)$ transforms the flow in the $z$-plane to a flow in the $w$-plane: if $F(z)$ is the complex potential for flow past an obstacle with boundary $\gamma$, composing with $\phi$ transforms the problem to flow past the image boundary $\phi(\gamma)$. This technique turns flow past a Joukowski airfoil (a complicated curve) into flow past a circle, which is elementary.
**Prime number theory.** The Riemann zeta function is defined for $\operatorname{Re}(s) > 1$ by the Dirichlet series
\begin{align*}
\zeta(s) &= \sum_{n=1}^\infty \frac{1}{n^s}.
\end{align*}
This series converges absolutely and defines a holomorphic function in the half-plane $\operatorname{Re}(s) > 1$. The Euler product formula $\zeta(s) = \prod_p (1 - p^{-s})^{-1}$ (product over primes $p$) encodes the fundamental theorem of arithmetic. Analytic continuation extends $\zeta$ to a meromorphic function on all of $\mathbb{C}$, with a simple pole at $s = 1$ and no other singularities. The prime number theorem — that the number of primes up to $x$ is asymptotic to $x/\ln x$ — is equivalent to the statement that $\zeta(s) \ne 0$ on the line $\operatorname{Re}(s) = 1$. The Riemann Hypothesis asserts the stronger statement that all non-trivial zeros (those in the critical strip $0 < \operatorname{Re}(s) < 1$) lie on the critical line $\operatorname{Re}(s) = 1/2$, and would give sharp error bounds for the prime counting function. It remains one of the central open problems in mathematics.
## Liouville's Theorem and the Fundamental Theorem of Algebra
The compactness of the Riemann sphere, the rigidity of holomorphic functions, and the Cauchy estimates combine to yield one of the most elegant corollaries in mathematics: a bounded entire function must be constant. This is Liouville's theorem, and its proof is a single application of the Cauchy estimates.
An **entire function** is a function holomorphic on all of $\mathbb{C}$, written $f \in \mathcal{O}(\mathbb{C})$. The polynomial $z^n$, the exponential $e^z$, the sine and cosine functions $\sin z$ and $\cos z$, and all convergent power series with infinite radius of convergence are entire. The striking feature of entire functions is that they are either unbounded or constant — there is nothing in between.
[quotetheorem:346]
The proof uses the Cauchy estimate for $n = 1$: for any $z_0 \in \mathbb{C}$ and any $r > 0$,
\begin{align*}
|f'(z_0)| &\le \frac{M}{r}.
\end{align*}
Since $r$ can be taken arbitrarily large (as $f$ is entire), we get $|f'(z_0)| = 0$ for all $z_0$. A holomorphic function with zero derivative on a connected set is constant.
Liouville's theorem provides an elegant proof of the Fundamental Theorem of Algebra, a result that ostensibly belongs to algebra but whose cleanest proof is complex-analytic.
[quotetheorem:347]
Suppose $p$ has no roots. Then $f(z) = 1/p(z)$ is entire. Since $|p(z)| \to \infty$ as $|z| \to \infty$, we have $|f(z)| \to 0$, so $f$ is bounded outside some large ball $|z| > R$. On the compact set $\overline{B}(0, R)$, $f$ is continuous and hence bounded. So $f$ is a bounded entire function. By Liouville's theorem, $f$ is constant — contradicting that $p$ is non-constant. The contradiction shows $p$ must have a root.
## The Identity Principle and Analytic Continuation
We now arrive at one of the most striking rigidity results in all of mathematics: a holomorphic function on a connected open set is completely determined by its values on any subset with an accumulation point.
To appreciate this, compare the real case. Define $g: \mathbb{R} \to \mathbb{R}$ to be $0$ on the interval $(-1, 1)$ and non-zero elsewhere, say $g(x) = (x-1)^2(x+1)^2$ for $|x| > 1$ smoothed appropriately. Such a construction is impossible for holomorphic functions. Once we know $f$ is zero on any sequence converging to a point inside the domain, $f$ is zero everywhere on the connected component.
[quotetheorem:208]
The proof reduces to showing that if $h = f - g$ vanishes on a set $S$ with an accumulation point $z_0 \in \Omega$, then the zero set of $h$ is both open and closed in $\Omega$. Since $\Omega$ is connected, the zero set must be either empty or all of $\Omega$. Openness follows from the power series expansion: if $h(z_0) = 0$ and $z_0$ is an accumulation point of zeros, then all Taylor coefficients $h^{(n)}(z_0)/n!$ must vanish (else $h$ could not vanish at nearby points by the theory of power series), so $h \equiv 0$ near $z_0$.
The identity theorem makes the theory of **analytic continuation** possible. If $f: \Omega_1 \to \mathbb{C}$ and $g: \Omega_2 \to \mathbb{C}$ are holomorphic and agree on $\Omega_1 \cap \Omega_2$ (which is nonempty and connected), then there is a unique holomorphic function on $\Omega_1 \cup \Omega_2$ extending both. This process can sometimes be continued along chains of overlapping domains, but may encounter obstructions — natural boundaries beyond which no extension is possible.
[example: The Gamma Function as Analytic Continuation]
The Euler integral $\Gamma(z) = \int_0^\infty t^{z-1} e^{-t}\, dt$ converges absolutely for $\operatorname{Re}(z) > 0$ and defines a holomorphic function there. Integration by parts yields the functional equation $\Gamma(z+1) = z\Gamma(z)$. This equation can be rewritten as
\begin{align*}
\Gamma(z) &= \frac{\Gamma(z+1)}{z},
\end{align*}
which extends $\Gamma$ to a holomorphic function on $\operatorname{Re}(z) > -1$ except at $z = 0$, where it has a simple pole with residue $\Gamma(1) = 1$. Iterating this argument $n$ times, we obtain the meromorphic extension to all of $\mathbb{C}$, with simple poles at $z = 0, -1, -2, \ldots$ and residue $(-1)^n / n!$ at $z = -n$.
The identity theorem guarantees that this extension is unique: any two holomorphic functions on $\mathbb{C} \setminus \{0, -1, -2, \ldots\}$ that agree with the Euler integral on $\operatorname{Re}(z) > 0$ must agree everywhere on their common domain.
[/example]
## The Maximum Modulus Principle
So far, the rigidity of holomorphic functions has manifested as restrictions on where zeros can accumulate and what bounded entire functions can look like. The maximum modulus principle expresses a different kind of rigidity: the maximum of $|f|$ on a closed region is always attained on the boundary, never in the interior.
This principle has a clean physical interpretation. Holomorphic functions are locally the real parts of harmonic functions (functions satisfying $\Delta u = 0$), and harmonic functions satisfy the mean value property: the value at any point equals the average over any surrounding circle. A function satisfying the mean value property cannot have an interior maximum without being constant, because any local maximum would have to equal the average of its values on a surrounding circle, which forces the surrounding values to all equal the maximum too — and by connectedness, the function is constant.
[quotetheorem:3345]
The hypothesis that $f$ is holomorphic is essential — real-valued smooth functions can attain interior maxima without being constant. The following example illustrates the contrast.
[example: Failure Without Holomorphicity]
The function $g(x, y) = x^2 + y^2$ on the unit disk $\Omega = \{(x,y) : x^2 + y^2 < 1\}$ attains its maximum on the boundary (where $g = 1$) and its minimum in the interior (at the origin, where $g = 0$). But the function $h(x, y) = 1 - x^2 - y^2$ attains its maximum at the interior point $(0, 0)$, with $h(0,0) = 1 > 0 = h|_{\partial\Omega}$. This is perfectly consistent: $h$ is a smooth real function, not a holomorphic one, and real smooth functions have no such constraint on their maxima.
Now consider $f(z) = z^2$ on the unit disk. The modulus $|f(z)| = |z|^2$ satisfies $|f(z)| \le 1$ for $|z| \le 1$, with equality if and only if $|z| = 1$. So the maximum of $|f|$ is attained on the boundary only. This is a special case of the maximum modulus principle.
[/example]
The maximum modulus principle constrains holomorphic functions on a bounded domain relative to their boundary values. A natural next question is: what can be said about a holomorphic map of the unit disk into itself that fixes the origin? The Schwarz lemma gives a sharp answer — such a map is contracting, and is a rotation if and only if the contraction is an equality.
[quotetheorem:368]
The Schwarz lemma is proved by applying the maximum modulus principle to $g(z) = f(z)/z$, which extends holomorphically to $z = 0$ (by Riemann's removable singularity theorem, since $g$ is bounded near $0$). The maximum modulus principle on $|z| \le r < 1$ gives $|g(z)| \le 1/r$ for $|z| \le r$; sending $r \to 1$ gives $|g(z)| \le 1$, i.e., $|f(z)| \le |z|$.
## Harmonic Functions and the Poisson Integral
The relationship between holomorphic functions and harmonic functions runs deep. If $f = u + iv \in \mathcal{O}(\Omega)$, then both $u$ and $v$ satisfy Laplace's equation:
\begin{align*}
\Delta u = \partial_x^2 u + \partial_y^2 u = 0, \qquad \Delta v = \partial_x^2 v + \partial_y^2 v = 0.
\end{align*}
This follows directly from the Cauchy–Riemann equations: differentiating $\partial_x u = \partial_y v$ with respect to $x$ and $\partial_y u = -\partial_x v$ with respect to $y$, then adding, gives $\Delta u = \partial_x \partial_y v - \partial_y \partial_x v = 0$ (mixed partials commute since holomorphic functions are $C^\infty$).
This means that Laplace's equation — the PDE governing steady-state heat distributions, electrostatic potentials, and irrotational fluid velocity fields — is intimately tied to complex analysis. But the connection goes further: every real solution of Laplace's equation on a simply connected domain is the real part of some holomorphic function, which means the full toolkit of complex analysis (power series, contour integrals, conformal mappings) is available for studying harmonic functions. This motivates isolating the class of solutions to Laplace's equation as a concept in its own right, both because these functions arise independently in physics and PDE, and because holomorphic function theory gives the most powerful tools for understanding them.
[definition: Harmonic Function]
Let $\Omega \subset \mathbb{R}^2$ be open. A function $u \in C^2(\Omega; \mathbb{R})$ is **harmonic** if
\begin{align*}
\Delta u &= \partial_x^2 u + \partial_y^2 u = 0 \quad \text{on } \Omega.
\end{align*}
[/definition]
Harmonic functions are exactly the real and imaginary parts of holomorphic functions. The real and imaginary parts of any holomorphic function are harmonic. The converse holds locally: if $u$ is harmonic on a simply connected domain $\Omega$, then there exists a harmonic function $v$ on $\Omega$ (called a **harmonic conjugate** of $u$) such that $f = u + iv$ is holomorphic on $\Omega$. The conjugate $v$ is determined up to an additive constant by the Cauchy–Riemann equations: $\partial_x v = -\partial_y u$ and $\partial_y v = \partial_x u$.
[remark: Global vs Local Existence of Harmonic Conjugates]
The "simply connected" hypothesis is essential. On $\Omega = \mathbb{C} \setminus \{0\}$, the function $u(x, y) = \ln\sqrt{x^2 + y^2} = \operatorname{Re}(\operatorname{Log} z)$ is harmonic. Its harmonic conjugate $v$ would satisfy $\nabla v = (-y/(x^2+y^2), x/(x^2+y^2))$, giving $v = \operatorname{Arg}(z)$. But the argument function is not single-valued on $\mathbb{C} \setminus \{0\}$: traversing a loop around the origin changes $\operatorname{Arg}(z)$ by $2\pi$. So $u$ has no global harmonic conjugate on $\mathbb{C} \setminus \{0\}$, even though it is harmonic everywhere there.
[/remark]
The mean value property for harmonic functions — which is itself a consequence of the holomorphic structure — provides a complete characterization: a continuous function is harmonic if and only if it satisfies the mean value property.
[quotetheorem:3346]
This follows directly from the Cauchy integral formula by substituting $z = z_0 + re^{i\theta}$, $dz = ire^{i\theta}\, d\theta$:
\begin{align*}
f(z_0) &= \frac{1}{2\pi i} \oint_{|z - z_0| = r} \frac{f(z)}{z - z_0}\, dz = \frac{1}{2\pi i} \int_0^{2\pi} \frac{f(z_0 + re^{i\theta})}{re^{i\theta}} \cdot ire^{i\theta}\, d\theta = \frac{1}{2\pi} \int_0^{2\pi} f(z_0 + re^{i\theta})\, d\theta.
\end{align*}
The mean value property tells us the value at the center of a disk is the average over the boundary circle. The Poisson integral formula refines this: it recovers the value at any interior point from boundary data on the circle, not just at the center. This is crucial for the boundary value problem: given continuous data $h$ prescribed on the boundary circle $\partial B(z_0, r)$, there is a unique harmonic function on $B(z_0, r)$ extending $h$, and the Poisson formula gives it explicitly.
[quotetheorem:1195]
The Poisson kernel $P_r(\theta, z)$ has a direct interpretation: it is a weighted average that assigns more weight to boundary points $re^{i\theta}$ that are geometrically close to $z$. When $z \to re^{i\varphi}$ (the point approaches the boundary), the kernel concentrates as a spike at $\theta = \varphi$, ensuring the boundary values are recovered continuously. This is the harmonic function analog of the Cauchy integral formula — and it can itself be derived from the Cauchy formula by taking real parts and using the formula for the real part of $(re^{i\theta} + z)/(re^{i\theta} - z)$.
## References
- L. Ahlfors, *Complex Analysis* (3rd ed., 1979). The classical graduate text; the geometric perspective on Möbius transformations and Riemann surfaces is unmatched.
- E.M. Stein and R. Shakarchi, *Complex Analysis* (Princeton Lectures in Analysis II, 2003). Exceptionally clear treatment; the approach through the Cauchy integral is particularly well-motivated.
- W. Rudin, *Real and Complex Analysis* (3rd ed., 1987). Develops real and complex analysis in parallel; the treatment of Hardy spaces connects holomorphic function theory to Fourier analysis.
- J.B. Conway, *Functions of One Complex Variable* (2nd ed., 1978). Thorough reference with detailed treatment of conformal mappings and analytic continuation.
- S. Lang, *Complex Analysis* (4th ed., 1999). Covers the main theory with careful attention to the role of topology.