Multivalued Functions and Branch Cuts

Multivalued functions are relations like

z

lo g z

that assign more than one output to a single complex input. Branch cuts are curves removed from the plane to force a single-valued, continuous branch. Key result: a function is multivalued precisely when it fails to return to its original value after analytic continuation around a branch point. Applies throughout complex analysis, contour integration, fluid dynamics, and quantum field theory. Concepts crystallized by Cauchy, Riemann, and Puiseux in the 19th century.

A multivalued function $w = f (z)$ on the complex plane is a rule that may associate several values $w$ to one $z$ ; the standard examples are the $n$ th root $z^{1/ n}$ , the complex logarithm $lo g z$ , and inverse trigonometric functions. They arise inescapably because the complex exponential and power maps are not injective — $e^{z}$ has period $2 π i$ , so its inverse $lo g z$ acquires an additive ambiguity $2 π ik$ . To recover an ordinary single-valued analytic function one cuts the plane along a branch cut anchored at branch points, selecting a branch of the function. The surprising fact is that this messiness is not a defect but a window onto deep geometry: the "correct" domain is not the plane at all but a multi-sheeted Riemann surface on which the function becomes perfectly single-valued and smooth.

The Core Insight

A function is multivalued exactly when transporting a value continuously around a closed loop fails to bring it home. The obstruction lives at isolated branch points. Removing arcs (branch cuts) that join branch points kills the offending loops, restoring single-valuedness — at the cost of introducing a discontinuity across the cut. The deletion is a choice; the function itself is honestly defined on a Riemann surface where no choice is needed.

Formal Statement

Multivalued Function and Branch

A multivalued function is a correspondence $F$ assigning to each $z$ in a domain $D \subseteq C$ a nonempty set $F (z) \subseteq C$ . A branch of $F$ on an open connected set $U \subseteq D$ is a single-valued analytic function $f : U \to C$ with $f (z) \in F (z)$ for all $z \in U$ .

The two archetypes:

lo g z = ln ∣ z ∣ + i ar g z = ln ∣ z ∣ + i (Arg z + 2 π k), k \in Z

⟳Invert

e^{w} = z

; write

w = u + i v

, so

e^{u} = ∣ z ∣

and

v \equiv ar g z (mod 2 π)

, giving

u = ln ∣ z ∣

and the additive

2 π k

z^{1/ n} = ∣ z ∣^{1/ n} exp (i \frac{Arg z + 2 π k}{n}), k = 0, 1, \dots, n - 1

⟳Solve

w^{n} = z

in polar form

w = ρ e^{i θ}

; match moduli

ρ^{n} = ∣ z ∣

and arguments

n θ \equiv ar g z (mod 2 π)

, yielding

n

distinct

θ

Here $Arg z \in (- π, π]$ denotes the principal argument, $ar g z$ the full multivalued argument, and $ln$ the real natural logarithm of the positive number $∣ z ∣$ .

Branch Point and Branch Cut

A point $z_{0}$ is a branch point of $F$ if, for every sufficiently small loop $γ$ encircling $z_{0}$ , analytic continuation of a branch around $γ$ does not return to its starting value. A branch point has order $m$ if $m$ encirclements (not fewer) restore the value; if no finite number does, the branch point is logarithmic.

A branch cut is a curve (often an arc or ray) $Γ \subseteq C$ joining branch points (possibly through $\infty$ ) such that $C ∖ Γ$ is simply connected and a single-valued analytic branch exists on it.

Monodromy / Existence of Branches

Let $F$ be an analytic multivalued function whose branch points form a discrete set $B$ . On any simply connected open set $U \subseteq C ∖ B$ , every local branch extends to a single-valued analytic branch on all of $U$ . Conversely, on a domain that is not simply connected and encircles a branch point, no global single-valued branch exists.

of the Monodromy Statement

Step 1. Fix

z_{1} \in U

and a local analytic branch

f_{1}

near

z_{1}

(it exists since

F

is analytic and

z_{1} \in / B

). Step 2. For any

z \in U

, choose a path

σ

from

z_{1}

z

lying in

U

. By analytic continuation along

σ

we obtain a value

f_{σ} (z)

. Step 3. Take a second path

σ^{'}

from

z_{1}

z

. Since

U

is simply connected,

σ

and

σ^{'}

are homotopic in

U

. Because

U \cap B = \emptyset

, the continuation is unchanged under homotopy (no branch point is crossed):

f_{σ} (z) = f_{σ^{'}} (z) for all paths σ, σ^{'} \subseteq U from z_{1} to z .

⟳The set of points where two continuations along homotopic paths agree is open and closed in the homotopy parameter; connectedness of the homotopy forces global agreement.

Step 4. Hence

f (z) := f_{σ} (z)

is well-defined independent of the path, single-valued, and analytic (being locally a continuation of an analytic germ). Step 5 (converse). If a loop

γ \subseteq U

winds once around a branch point

z_{0}

, then by definition continuation around

γ

changes the value, so no single-valued continuous

f

can satisfy

f \in F

on all of

U

■

Bernhard Riemann | 1826–1866

Riemann's 1851 doctoral dissertation and his 1857 work on Abelian functions reconceived multivalued functions geometrically. Rather than mutilating the plane with arbitrary cuts, he proposed building a surface — now the Riemann surface — by gluing copies (sheets) of the plane along the cuts so that the function $z$ , $lo g z$ , or an algebraic function becomes genuinely single-valued and analytic. The branch points become ramification points where sheets meet. This shifted complex analysis from manipulating formulas to studying the topology of the natural domain, seeding modern algebraic geometry and topology.

Building Intuition

Picture $z = r e^{i θ}$ as a point on a turntable; its angle $θ$ is recorded on an unbounded tape, not a dial — once around adds $2 π$ to $θ$ , not back to zero. For $lo g z = ln r + i θ$ , this extra $2 π$ leaks straight into the imaginary part: walk a full circle around the origin and the logarithm jumps by $2 π i$ . The origin is the pivot you cannot circle without paying this toll — the branch point. A branch cut, say the negative real axis, is a fence you forbid yourself from crossing; with the fence in place you can never complete the toll-incurring loop, so $lo g z$ stays single-valued, jumping by $2 π i$ as you teleport across the fence.

For $z$ the toll is gentler: one loop multiplies $z$ by $e^{iπ} = - 1$ , but two loops give $e^{2 π i} = + 1$ and you're home. So the origin is an order-2 branch point — you need a 2-sheeted surface, not infinitely many sheets as for $lo g$ .

Branch cut for $\sqrt{z}$ along the negative real axis: crossing the cut flips the sign of the square root

Worked Examples

example[Principal logarithm of a complex number] Given: Compute $Log (- 1 + i)$ using the principal branch, where $Log z = ln ∣ z ∣ + i Arg z$ , $Arg z \in (- π, π]$ .

Step 1 — modulus:

∣ - 1 + i ∣ = (- 1)^{2} + 1^{2} = 2 .

⟳

∣ z ∣ = x^{2} + y^{2}

with

x = - 1, y = 1

Step 2 — principal argument. The point

(- 1, 1)

lies in the second quadrant:

Arg (- 1 + i) = π - arctan \frac{1}{1} = π - \frac{π}{4} = \frac{3 π}{4} .

⟳For

x < 0, y > 0

Arg = π - arctan (∣ y ∣/∣ x ∣)

Step 3 — assemble:

Log (- 1 + i) = ln 2 + i \frac{3 π}{4} = \frac{1}{2} ln 2 + i \frac{3 π}{4} .

⟳Substitute Steps 1–2 into

Log z = ln ∣ z ∣ + i Arg z

;

ln 2 = \frac{1}{2} ln 2

Log (- 1 + i) = 0.3466 + 2.3562 i (dimensionless)

Other branches add $2 π ik$ : e.g. $k = 1$ gives $0.3466 + 8.6394 i$ .

All cube roots and choosing a branch

Given: Find all solutions of

w^{3} = - 8 i

, then identify which one belongs to the principal branch of

z^{1/3}

. Step 1 — polar form of $z = - 8 i$ :

∣ z ∣ = 8

Arg z = - π /2

(negative imaginary axis).

z = 8 exp (i (- \frac{π}{2} + 2 π k)) .

⟳

- 8 i

has modulus

8

and lies on the negative imaginary axis, so

Arg = - π /2

Step 2 — apply the root formula

[2]

with

n = 3

w_{k} = 8^{1/3} exp (i \frac{- π /2 + 2 π k}{3}) = 2 exp (i \frac{- π + 4 π k}{6}), k = 0, 1, 2.

⟳

8^{1/3} = 2

; combine the angle into a single fraction over

6

Step 3 — evaluate:

$k = 0$ : angle $- π /6 \Rightarrow w_{0} = 2 (cos (- 3 0^{\circ}) + i sin (- 3 0^{\circ})) = 3 - i .$
$k = 1$ : angle $π /2 \Rightarrow w_{1} = 2 i .$
$k = 2$ : angle $7 π /6 \Rightarrow w_{2} = - 3 - i .$

Step 4 — principal branch uses

k = 0

(since it keeps

Arg z /3 \in (- π /3, π /3]

(- 8 i)_{principal}^{1/3} = 3 - i \approx 1.732 - 1.000 i (dimensionless)

Check: $(3 - i)^{3} = (3)^{3} - 3 (3)^{2} i + 33 i^{2} - i^{3} = 33 - 9 i - 33 + i = - 8 i .$ ✓

Common Misconceptions

The identity $lo g (z_{1} z_{2}) = lo g z_{1} + lo g z_{2}$ does NOT hold for the principal branch. It holds only modulo

2 π i

, or as a set equality among all branches. For example, with

z_{1} = z_{2} = - 1

Log ((- 1) (- 1)) = Log (1) = 0, but Log (- 1) + Log (- 1) = iπ + iπ = 2 π i \neq = 0.

Similarly,

z_{1} z_{2} = z_{1} z_{2}

and

(z^{a})^{b} = z^{ab}

can fail for principal branches. Always check argument ranges before manipulating.

Branch cut location matters

A branch cut is not unique: any curve from the branch point to infinity (or between branch points) that disconnects the bad loops works. For $lo g z$ one may cut along the negative real axis (principal), the positive real axis, or any ray from $0$ to $\infty$ . These give genuinely different single-valued functions. Consider evaluating just below the negative real axis at $z = - 1 - ε i$ :

Principal cut (negative real axis): the point sits on the cut; approaching from below gives $ar g \to - π$ .
Positive-axis cut ( $ar g \in (0, 2 π)$ ): same point now has $ar g \approx π$ , fully analytic there.

The numerical value of " $lo g (- 1)$ " thus depends entirely on the chosen cut — there is no branch-independent answer.

Connections

Multivalued functions are the inverse-image manifestation of non-injective analytic maps, tying directly into Complex Exponential and Logarithm and the failure of global inverses. The need to delete cuts to obtain analyticity links to Cauchy-Riemann Equations (which a chosen branch must satisfy) and to Contour Integration, where branch cuts force keyhole contours and contribute jump discontinuities across the cut. The clean resolution lives in Riemann Surfaces, where the multi-sheeted domain makes the function single-valued; the topology of these surfaces connects to Fundamental Group and covering-space theory. Finally, the residue-free contribution of a branch cut to integrals connects to Residue Theorem generalizations.

Real-World Applications

Aerodynamics — Joukowski transform. The conformal map $w = z + 1/ z$ , built from multivalued inverses, maps a circle to an airfoil; branch behavior of its inverse governs the flow's circulation and lift in classical wing theory.
Electrostatics & 2D fluid flow. The complex potential $Ω (z) = \frac{Γ}{2 π i} lo g (z - z_{0})$ for a vortex/line charge is intrinsically multivalued; the $2 π i$ jump encodes total circulation or enclosed charge.
Quantum field theory — propagators. Feynman propagators and dispersion relations contain logarithms and square roots of $s = p^{2}$ ; branch cuts along thresholds (e.g. $s \geq 4 m^{2}$ ) encode particle-production channels and the optical theorem.
Signal processing — phase unwrapping. Computing the phase $ar g H (e^{iω})$ of a transfer function requires "unwrapping" the principal-branch jumps of $2 π$ — a direct engineering encounter with multivaluedness.
Geophysics & seismology. Square-root branch points of the slowness/dispersion function determine wave-mode cuts in the complex wavenumber plane, controlling surface (Rayleigh) wave existence.
Conformal mapping in heat conduction. Temperature fields around corners and slits use $z^{α}$ with non-integer $α$ , whose branch cut models the physical edge of a conductor.
Control theory — Nyquist/Bode analysis. The argument principle counts encirclements using $lo g$ of the loop transfer function; branch behavior underlies stability margins.

Historical Development

1748 — Euler publishes $e^{i θ} = cos θ + i sin θ$ , implicitly revealing the periodicity that makes $lo g$ multivalued.
1714/1747 — A long dispute (Leibniz, Johann Bernoulli, then resolved by Euler) over the value of $lo g (- 1)$ exposes the multivaluedness of the logarithm of negatives.
1825 — Cauchy develops contour integration and begins systematic treatment of integrals over paths avoiding singularities, the seed of branch-cut handling.
1851 — Riemann's dissertation introduces Riemann surfaces, giving multivalued functions a single-valued home and recasting branch points as ramification.
1850s — Victor Puiseux analyzes algebraic functions near branch points, producing Puiseux series (fractional-power expansions) and clarifying branch-point order.
1857 — Riemann's Theorie der Abel'schen Functionen connects branch points to genus and topology of the surface.
1880s — Weierstrass formalizes analytic continuation via power-series germs, making rigorous the notion of "carrying a value around a loop."
1913 — Hermann Weyl's Die Idee der Riemannschen Fläche gives the first rigorous abstract definition of a Riemann surface, decoupling it from embeddings in the plane.
20th c. — Branch cuts become standard tools in physics (Sommerfeld, Watson) for evaluating diffraction and dispersion integrals.

Subtleties & Limits

While the existence of Riemann surfaces resolving classical multivalued functions is fully understood, several frontiers remain active and subtle. For general analytic functions the global monodromy group — the set of permutations of branches induced by loops — can be enormously complex; computing or even bounding it for specific families (e.g. solutions of Painlevé equations or hypergeometric functions with many singular points) is hard. Resurgence theory and the modern study of Stokes phenomena treat how branch-cut data and exponentially small terms encode hidden saddle contributions in asymptotic series — a domain bridging Écalle's alien calculus and physics, still incompletely systematized. In several complex variables, branch loci become codimension-one varieties and the simple "cut a curve" picture fails; the structure of ramification for multivariate algebraic functions and the topology of their normalizations remain technically demanding. Finally, the algorithmic problem of choosing globally consistent branch cuts for compositions of special functions (as in symbolic computation systems) has no universally agreed convention and produces genuine inconsistencies across software.