Complex analysis · Foundations

Complex Functions: Limits, Continuity and the Elementary Functions

A function of a complex variable takes a point of one plane and sends it to a point of another, so it is best thought of as a mapping rather than a graph. Before we can do calculus with such mappings we need a vocabulary for sets in the plane (neighbourhoods, domains, curves) and a careful look at the elementary functions, some of which turn out to have several values at once. This post builds that vocabulary and ends with limits and continuity, which look exactly like their real versions but with \(|\cdot|\) meaning distance in the plane.

1. Topological notions in the complex plane

Everything rests on distance: \(|z_1 - z_2|\) is the ordinary distance between the points \(z_1\) and \(z_2\) of the plane. "Close to \(z_0\)" means "inside a small disk around \(z_0\)".

Neighbourhoods The disk (open disk) of radius \(\varepsilon\) around \(z_0\) is \[ V_{z_0}(\varepsilon) = \{z : |z - z_0| < \varepsilon\}, \qquad \varepsilon > 0. \] A neighbourhood of \(z_0\) is any set containing such a disk. A neighbourhood of the point \(z = \infty\) is any set containing the exterior of a circle, \( \{z : |z| > R\} \), \(R > 0\).

The point \(\infty\) is added to the plane so that "\(z\) goes far away in any direction" becomes "\(z\) approaches one point". The resulting extended complex plane \( K \cup \{\infty\} \) can be pictured as a sphere (the Riemann sphere), with \(\infty\) at the north pole: large circles \(|z| = R\) become small circles around that pole.

Let \(E\) be a set of points. With disks, the following notions are defined:

NotionDefinitionIntuition
Accumulation point of \(E\)every neighbourhood of \(z_0\) contains a point \(z_1 \ne z_0\) of \(E\)points of \(E\) crowd arbitrarily close to \(z_0\)
Closed setcontains all its accumulation pointsyou cannot "escape" \(E\) by a limit
Interior pointsome disk around \(z_0\) lies in \(E\)\(z_0\) has room in all directions
Boundary pointevery disk around \(z_0\) contains points of \(E\) and points not in \(E\)on the edge
Open setconsists only of interior pointsno edge points included
Bounded set\( |z| < M \) for all \(z \in E\), some \(M > 0\)fits in a disk
Example 1 For the disk \( |z| < 1 \): every point is interior, so it is open; the circle \( |z| = 1 \) consists of boundary points, which are accumulation points not belonging to the set, so it is not closed. The set \( |z| \le 1 \) is closed and bounded. The set \( \{1, \frac12, \frac13, \dots\} \) has the single accumulation point \(0\), which does not belong to it.

2. Curves and domains

Curves
  • A continuous curve \( z = z(t) = x(t) + iy(t) \), \( t \in [a, b] \), is a Jordan curve if it never passes through the same point twice: \( t_1 \ne t_2 \Rightarrow z(t_1) \ne z(t_2) \). It is a closed Jordan curve if \( z(a) = z(b) \), with no other self-intersections.
  • It is smooth if \(\dot x(t)\), \(\dot y(t)\) are continuous and \( \dot x^2(t) + \dot y^2(t) > 0 \) on \([a, b]\): there is a tangent everywhere, and it turns continuously.
  • A curve \(\Gamma\) made of finitely many smooth pieces is a contour (path). Think of the boundary of a square: four smooth sides, corners allowed.

The condition \( \dot x^2 + \dot y^2 > 0 \) excludes moments where the moving point stops; at such moments the curve can form a sharp cusp even though \(x(t)\) and \(y(t)\) are smooth.

Domains
  • An open set \(G\) is a domain (region) if any two of its points can be joined by a polygonal line lying entirely in \(G\): it is "in one piece".
  • \( \overline G = G \cup \{\text{boundary points of } G\} \) is a closed domain.
  • \(G\) is simply connected (in the finite plane) if for every closed Jordan curve \(\Gamma \subset G\), the interior of \(\Gamma\) also lies in \(G\): \( \operatorname{int}\Gamma \subset G \). Otherwise it is multiply connected.

Intuitively, a simply connected domain has no holes: every loop in it can be shrunk to a point without leaving the domain. An annulus \( 1 < |z| < 2 \) is not simply connected: a loop around the hole encloses points (\(z = 0\)) that are not in the domain. This distinction will be crucial for Cauchy's integral theorem.

z₁ z₂ E interior / boundary Γ simply connected Γ hole doubly connected
Figure 1. Left: a set \(E\) with an interior point \(z_1\) (a whole disk fits inside) and a boundary point \(z_2\) (every disk sticks out). Middle: a simply connected domain; every closed curve encloses only points of the domain. Right: a doubly connected domain (one hole); the curve \(\Gamma\) encloses points that are not in the domain.

In the extended plane, the definition is adjusted: \(G\) is simply connected if for each closed Jordan curve \(\Gamma \subset G\), either the interior or the exterior of \(\Gamma\) lies in \(G\). For example, the exterior \(|z| > 1\) together with \(\infty\) is simply connected on the Riemann sphere (it is a disk around the north pole).

3. A complex function is a mapping

Complex function Let \(E_z\) be a set of points of the \(z\)-plane and \(E_w\) a set of the \(w\)-plane. A function \( f: E_z \to E_w \) is called a function of a complex variable, or a mapping of \(E_z\) into \(E_w\): \[ w = f(z) = u(x, y) + iv(x, y), \qquad z = x + iy, \] where \(u\) and \(v\) are real functions of two real variables. If one value \(z\) corresponds to several values \(w\), \(f\) is a multivalued function.

A graph of \(f\) would need four real dimensions (two for \(z\), two for \(w\)), so instead we draw two planes and watch how curves and regions of the \(z\)-plane are transported to the \(w\)-plane. Every complex function is a pair of real functions \(u, v\), but not every such pair is interesting: the special pairs that make \(f\) differentiable (the Cauchy–Riemann equations) are the subject of the next section of the course.

Example 2 · \(w = z^2\) \( (x + iy)^2 = x^2 - y^2 + 2ixy \), so \( u = x^2 - y^2 \), \( v = 2xy \). In polar form \( z = re^{i\varphi} \mapsto w = r^2e^{2i\varphi} \): distances from \(0\) are squared and angles are doubled. The first quadrant (\(0 \le \varphi \le \frac\pi2\)) is mapped onto the whole upper half-plane (\(0 \le \arg w \le \pi\)).

4. The basic elementary functions

FunctionDefinitionDefined for
Power\( w = z^n \), \( n \in \mathbb N \)all \(z\)
Polynomial\( P_n(z) = a_nz^n + a_{n-1}z^{n-1} + \cdots + a_1z + a_0 \)all \(z\)
Rational\( w = \frac{P_n(z)}{Q_m(z)} \)\(z\) not a zero of \(Q_m\)
Exponential\( e^z = e^x(\cos y + i\sin y) \)all \(z\)
Sine, cosine\( \sin z = \frac{e^{iz} - e^{-iz}}{2i} \), \( \cos z = \frac{e^{iz} + e^{-iz}}{2} \)all \(z\)
Tangent, cotangent\( \tan z = \frac{\sin z}{\cos z} \), \( \cot z = \frac{\cos z}{\sin z} \)\( z \ne (k + \frac12)\pi \); \( z \ne k\pi \) (\(k \in \mathbb Z\))
Hyperbolic\( \sinh z = \frac{e^z - e^{-z}}{2} \), \( \cosh z = \frac{e^z + e^{-z}}{2} \), \( \tanh z \), \( \coth z \)\(\cosh z \ne 0\), resp. \(\sinh z \ne 0\)

The definitions are chosen so that for real \(z\) everything agrees with the familiar real functions, and the familiar identities (\( e^{z_1 + z_2} = e^{z_1}e^{z_2} \), \( \sin^2 z + \cos^2 z = 1 \), addition formulas) still hold. But some properties change dramatically:

Example 3 · Surprises
  • \(e^z\) is periodic with period \(2\pi i\): \( e^{z + 2\pi i} = e^z \), because \(\cos y\) and \(\sin y\) are \(2\pi\)-periodic. It is never zero, since \( |e^z| = e^x > 0 \).
  • Sine and cosine are unbounded. \( \cos(iy) = \frac{e^{-y} + e^y}{2} = \cosh y \) and \( \sin(iy) = i\sinh y \). For example \( \cos i = \cosh 1 \approx 1.543 > 1 \). The inequality \(|\cos z| \le 1\) holds only on the real axis.
  • Trigonometric and hyperbolic functions are the same family: \( \cos(iz) = \cosh z \), \( \sin(iz) = i\sinh z \). The complex exponential unifies them.

5. Multivalued functions

Inverse functions of periodic or non-injective functions cannot be single-valued. In the complex plane this happens to the most basic inverses.

Roots

The inverse of \( w = z^n \) is \( w = \sqrt[n]{z} \), which has \(n\) values for \(z \ne 0\): with \( z = re^{i\varphi} \),

\[ \sqrt[n]{z} = \sqrt[n]{r}\,e^{i(\varphi + 2k\pi)/n}, \qquad k = 0, 1, \dots, n - 1, \]

the vertices of a regular \(n\)-gon.

The logarithm

Logarithm The inverse of \(w = e^z\) is \[ w = \operatorname{Ln}z = \ln z + 2k\pi i \qquad (k \in \mathbb Z,\ z \ne 0), \] where \( \ln z = \ln|z| + i\arg z \) (with the principal argument \( -\pi < \arg z \le \pi \)) is the single-valued principal value of \(\operatorname{Ln}z\).

Why: \( e^w = z \) with \( w = u + iv \) means \( e^u = |z| \) and \( v = \) any angle of \(z\). The modulus fixes \(u = \ln|z|\), but the angle is determined only up to multiples of \(2\pi\), which is the periodicity of \(e^z\) seen from the other side. So the logarithm has infinitely many values, spaced \(2\pi i\) apart.

Example 4 · Logarithms of negative numbers \( \operatorname{Ln}(-1) = \ln 1 + i\pi + 2k\pi i = (2k + 1)\pi i \); the principal value is \( \ln(-1) = i\pi \), which is Euler's \( e^{i\pi} = -1 \) read backwards. Also \( \operatorname{Ln}i = \frac{\pi i}{2} + 2k\pi i \).

General powers, exponentials and logarithms

With the logarithm, every power can be defined through the exponential:

\[ a^z = e^{z\operatorname{Ln}a} \ (a \ne 0), \qquad z^\lambda = e^{\lambda\operatorname{Ln}z}, \qquad \operatorname{Log}_az = \frac{\operatorname{Ln}z}{\operatorname{Ln}a} \ (a \ne 0, 1), \]

where in the last formula \(\operatorname{Ln}a\) means one chosen branch. All are generally multivalued.

Example 5 · \(i^i\) is real \[ i^i = e^{i\operatorname{Ln}i} = e^{i(\frac{\pi i}{2} + 2k\pi i)} = e^{-\frac\pi2 - 2k\pi}. \] Infinitely many values, all real and positive. The principal one (\(k = 0\)) is \( e^{-\pi/2} \approx 0.2079 \).

Inverse trigonometric functions

Solving \( z = \sin w = \frac{e^{iw} - e^{-iw}}{2i} \) for \(w\) is a quadratic equation in \(e^{iw}\): \( e^{2iw} - 2iz\,e^{iw} - 1 = 0 \), so \( e^{iw} = iz \pm\sqrt{1 - z^2} \). Hence

\[ \operatorname{Arcsin}z = \frac1i\operatorname{Ln}\big(iz \pm\sqrt{1 - z^2}\big), \qquad \operatorname{Arctan}z = \frac{1}{2i}\operatorname{Ln}\frac{1 + iz}{1 - iz} \ (z \ne \pm i). \]
Example 6 · \(\sin w = 2\) has solutions In the reals, \(\sin w = 2\) is impossible. In the complex plane: \[ \operatorname{Arcsin}2 = -i\operatorname{Ln}\big(2i \pm i\sqrt3\big) = -i\operatorname{Ln}\big(i(2 \pm\sqrt3)\big) = \frac\pi2 + 2k\pi \mp i\ln\big(2 + \sqrt3\big), \] using \( \ln(2 - \sqrt3) = -\ln(2 + \sqrt3) \). For example \( w = \frac\pi2 - 1.317i \): indeed \( \sin\big(\frac\pi2 - it\big) = \cosh t \), and \( \cosh(\ln(2 + \sqrt3)) = 2 \).

Algebraic functions

Algebraic function An algebraic function is a solution \(w\) of an equation \[ P_0(z)w^n + P_1(z)w^{n-1} + \cdots + P_{n-1}(z)w + P_n(z) = 0, \] where \( P_0(z) \not\equiv 0, P_1(z), \dots, P_n(z) \) are polynomials in \(z\) and \(n \in \mathbb N\).

For each \(z\) this is a polynomial equation of degree \(n\) in \(w\), so (generically) it has \(n\) solutions: an algebraic function is in general \(n\)-valued. Example: \( w^2 - z = 0 \) defines \( w = \sqrt z \).

6. Branch points

Branch point A point \(z_0\) is a branch point (critical point) of a multivalued function if, when \(z\) describes a closed Jordan curve around \(z_0\) (and the value of the function is followed continuously), the image curve in the \(w\)-plane does not close up.
  • If the image closes only after the curve goes around \(z_0\) exactly \(n\) times, \(z_0\) is an algebraic branch point of order \(n\).
  • If the image never closes, however many times we go around, \(z_0\) is a transcendental branch point.

Follow \( z = re^{i\varphi} \) once around the origin, \(\varphi\) from \(0\) to \(2\pi\):

z-plane w-plane r 0 √r −√r w = √z
Figure 2. The point \(z\) goes once around the circle \(|z| = r\) (left). The value of \(\sqrt z\), followed continuously, moves only along a half-circle of radius \(\sqrt r\) (right): it starts at \(\sqrt r\) and ends at \(-\sqrt r\). The image does not close, so \(z = 0\) is a branch point of \(\sqrt z\).

To work with single-valued functions anyway, one makes a cut (for example along the negative real axis) that curves are not allowed to cross. In the cut plane each choice of value (each branch) is a genuine single-valued, continuous function, e.g. the principal value \(\ln z\).

7. Limits

Limit A function \(w = f(z)\), defined in a neighbourhood of \(z_0\) except possibly at \(z_0\), has the limit \(A\) as \(z \to z_0\) if for every \(\varepsilon > 0\) there is \(\delta = \delta(\varepsilon) > 0\) such that \[ |f(z) - A| < \varepsilon \quad\text{for all } z \ne z_0 \text{ in the disk } V_{z_0}(\delta). \] Notation: \( \lim_{z\to z_0}f(z) = A \) or \( f(z) \to A \) (\(z \to z_0\)). For \(z_0 = \infty\), the disk is replaced by the exterior \( \{z : |z| > R\} \) of a (large) circle.

The definition reads exactly like the real one, but there is a big difference in what it demands. In the real case, \(x\) can approach \(x_0\) only from the left or the right. In the plane, \(z\) can approach \(z_0\) along infinitely many paths (lines, spirals, ...), and the limit must be the same along all of them, just as for functions of two real variables (see the post on limits in several variables).

Example 7 · A limit that does not exist \( f(z) = \frac{\bar z}{z} \) as \( z \to 0 \). Along the real axis (\(z = x\)): \( \frac xx = 1 \). Along the imaginary axis (\(z = iy\)): \( \frac{-iy}{iy} = -1 \). Two different values, so there is no limit. (In polar form \( \frac{\bar z}{z} = e^{-2i\varphi} \) depends only on the direction of approach.)
Limits of sums, products, quotients If \( f(z) \to A_1 \) and \( g(z) \to A_2 \) as \(z \to z_0\), then \[ f(z) \pm g(z) \to A_1 \pm A_2, \qquad f(z)g(z) \to A_1A_2, \qquad \frac{f(z)}{g(z)} \to \frac{A_1}{A_2} \ (A_2 \ne 0). \]

Also useful: \( f(z) \to A \) if and only if \( u \to \operatorname{Re}A \) and \( v \to \operatorname{Im}A \), since \( \max(|\Delta u|, |\Delta v|) \le |\Delta f| \le |\Delta u| + |\Delta v| \). For example, \( \lim_{z\to\infty}\frac1z = 0 \) and \( \lim_{z\to\infty}\frac{2z + 1}{z - i} = 2 \).

8. Continuity and uniform continuity

Continuity
  1. \(f\) is continuous at \(z_0\) if \( \lim_{z\to z_0}f(z) = f(z_0) \).
  2. \(f\) is continuous in a domain \(G\) if it is continuous at every point of \(G\).
  3. \(f = u + iv\) is continuous in \(G\) if and only if \(u(x, y)\) and \(v(x, y)\) are continuous in \(G\).
  4. A function continuous on a bounded closed domain is bounded there (and attains the maximum of \(|f|\)).
"Closed" is not enough Point 4 needs the domain to be both closed and bounded. The function \(f(z) = z\) is continuous on the whole plane, which is closed, but it is not bounded. (On a bounded closed set, compactness guarantees boundedness, as in Weierstrass's theorem for real functions on \([a, b]\).)

All the elementary functions of Section 4 are continuous wherever they are defined; branches of multivalued functions are continuous in a suitably cut plane (e.g. \(\ln z\) is continuous off the negative real axis, but jumps by \(2\pi i\) across it).

Uniform continuity \(f\) is uniformly continuous in \(G\) if for every \(\varepsilon > 0\) there is \(\delta = \delta(\varepsilon) > 0\) such that \[ |f(z_1) - f(z_2)| < \varepsilon \quad\text{whenever}\quad |z_1 - z_2| < \delta, \qquad z_1, z_2 \in G. \]

The difference from ordinary continuity: \(\delta\) must work for the whole domain at once, not just near one point. As in the real case, a function continuous on a bounded closed domain is uniformly continuous there (Cantor's theorem).

Example 8 \( f(z) = \frac1z \) is continuous in \( 0 < |z| < 1 \), but not uniformly: the points \( z_1 = \frac1n \), \( z_2 = \frac{1}{n + 1} \) are arbitrarily close for large \(n\), while \( |f(z_1) - f(z_2)| = 1 \). On the closed annulus \( \frac12 \le |z| \le 1 \) it is uniformly continuous (there \( |\frac1{z_1} - \frac1{z_2}| = \frac{|z_1 - z_2|}{|z_1||z_2|} \le 4|z_1 - z_2| \)).

9. Summary

ConceptKey point
Neighbourhoodcontains a disk \(|z - z_0| < \varepsilon\); of \(\infty\): contains \(|z| > R\)
Domainopen and connected (polygonal paths)
Simply connectedno holes: \(\operatorname{int}\Gamma \subset G\) for closed \(\Gamma \subset G\)
Complex functionmapping \(w = u(x, y) + iv(x, y)\)
\(e^z\), \(\sin z\), \(\cos z\)\(e^z\) has period \(2\pi i\); \(\sin\), \(\cos\) unbounded
\(\operatorname{Ln}z\)\( \ln|z| + i\arg z + 2k\pi i \): infinitely many values
Branch pointalgebraic of order \(n\) (\(\sqrt[n]z\)) or transcendental (\(\operatorname{Ln}z\))
Limitsame along all paths; \(f \to A \iff u, v\) converge
Continuity\(\lim f = f(z_0)\); bounded closed domain \(\Rightarrow\) bounded and uniformly continuous

10. Practice problems

Try each one before opening the solution. More solved problems from this topic are in Complex Numbers.

Problem 1: Find \(u\) and \(v\) for \( w = \frac1z \).

\( \frac1z = \frac{\bar z}{|z|^2} = \frac{x - iy}{x^2 + y^2} \): \( u = \frac{x}{x^2 + y^2} \), \( v = -\frac{y}{x^2 + y^2} \).

Problem 2: Compute all values of \(\sqrt[3]{-8}\).

\( -8 = 8e^{i\pi} \): \( 2e^{i(\pi + 2k\pi)/3} \), \(k = 0, 1, 2\): \( 1 + i\sqrt3 \), \(-2\), \( 1 - i\sqrt3 \).

Problem 3: Solve \( e^z = -2 \).

\( z = \operatorname{Ln}(-2) = \ln 2 + i\pi + 2k\pi i = \ln 2 + (2k + 1)\pi i \).

Problem 4: Show that \( |\sin z|^2 = \sin^2 x + \sinh^2 y \). Where is \(\sin z = 0\)?

\( \sin(x + iy) = \sin x\cosh y + i\cos x\sinh y \); then \( |\sin z|^2 = \sin^2x\cosh^2y + \cos^2x\sinh^2y = \sin^2 x + \sinh^2 y \) (using \(\cosh^2 = 1 + \sinh^2\)). It is zero only when \(\sin x = 0\) and \(y = 0\): \( z = k\pi \), the real zeros only.

Problem 5: Does \( \lim_{z\to0}\frac{\operatorname{Re}z}{z} \) exist?

Along the real axis: \(1\); along the imaginary axis: \(0\). No.

Problem 6: Is \( 1 < |z| < 2 \) a domain? Is it simply connected? What about \( |z| > 1 \) in the extended plane?

The annulus is open and connected, so a domain, but not simply connected (the circle \(|z| = 1.5\) encloses \(0\), which is not in it). \( \{|z| > 1\} \cup \{\infty\} \) is simply connected in the extended plane: for every closed curve in it, its exterior (the part containing \(\infty\)) lies in the domain.

Final thoughts

Moving from real to complex variables changes the scenery more than the definitions. Limits and continuity are defined with the same \(\varepsilon\)–\(\delta\) words, but approaching a point from infinitely many directions makes limits more demanding. The elementary functions keep their identities but lose some old certainties: \(e^z\) becomes periodic, \(\sin z\) and \(\cos z\) become unbounded, and the inverses (roots, logarithms, \(\operatorname{Arcsin}\)) become multivalued, with branch points where their values get tangled. The topological vocabulary (domains, contours, simple connectivity) is what the rest of complex analysis is built on, starting with derivatives and contour integrals.

Related posts: limits and continuity in several variables, and the Fourier integral (where \(e^{i\lambda x}\) is used constantly).