Branch Points, Branch Cuts, and Riemann Surfaces

Analytic function theory — built on the Cauchy–Riemann equations, contour integration, and Taylor/Laurent series — quietly assumes functions are single-valued. Yet the two most elementary inverse operations, taking roots and taking logarithms, violate that assumption at the root, because the exponential and power maps are not injective. Writing $z=re^{i\theta}$ , the angle $\theta$ is defined only modulo $2\pi$ , so $\log z=\ln r+i\theta$ carries an irreducible $2\pi i$ ambiguity, and $z^{1/2}=\sqrt r\,e^{i\theta/2}$ changes sign when $\theta$ advances by $2\pi$ . This article is about domesticating that ambiguity without pretending it away. Three ideas organize everything that follows. First, monodromy: continue a function around a closed loop and ask whether it returns to its starting value. A point some loop refuses to forgive is a branch point — for $\sqrt z$ they sit at $z=0$ and $z=\infty$ , and one circuit of either swaps the two roots. Second, the branch cut: a curve joining branch points that the path may not cross, carving a domain on which one single-valued branch lives — analytic except discontinuous across the cut, whose location is a free choice while its endpoints are not. Third, Riemann's cure — the Riemann surface — which abolishes cuts by enlarging the domain: stack one plane-copy per value, slit each along the cut, and cross-glue so that crossing a cut slides you smoothly onto the next sheet. On that connected surface the multivalued function becomes genuinely single-valued and analytic, and its topology (its genus) records the function's global anatomy. For physics the discontinuity across a cut is not a defect but data: the cut of a propagator in the complex-energy plane is the continuum of physical states, and the jump across it is the spectral density that unitarity ties to decay rates and cross sections — threading the geometry of Riemann surfaces straight through to measurable spectra.

Key takeaways

A multivalued function such as $\sqrt z$ or $\log z$ fails to return to its value when continued around a branch point; branch points are the intrinsic obstructions to a single-valued analytic branch.
A branch cut is a curve joining branch points (or running to infinity) that we forbid the path to cross, leaving a domain on which one single-valued, continuous branch exists — at the cost of a jump across the cut.
The location of a cut is arbitrary and movable; the existence and location of the branch points are intrinsic and fixed by where monodromy is nontrivial.
The intrinsic, cut-free cure is the Riemann surface: a multi-sheeted cover of the plane glued along the cuts, on which the multivalued function is a single genuine analytic function.
The jump across a cut, $\Delta f(x)=f(x+i0)-f(x-i0)$ , is itself a physical and mathematical object — it powers dispersion relations and the keyhole/dogbone evaluation of real integrals.
A function gains the factor $e^{2\pi i\alpha}$ around a branch point of order $\alpha$ : $\sqrt z$ needs two sheets, $z^{1/n}$ needs $n$ , and $\log z$ needs infinitely many.
Key idea: a branch point is a point around which a loop changes the function's value; a branch cut is a movable curtain joining branch points that restores single-valuedness at the price of a discontinuity; the Riemann surface is the canonical cure that removes the curtain by stacking and gluing sheets so the function is genuinely single-valued.

The core insight

A branch point is a point around which a loop changes a function's value; a branch cut is a movable curtain joining branch points that restores single-valuedness at the cost of a discontinuity; and the Riemann surface is the canonical cure that stacks and glues sheets so the multivalued function becomes genuinely single-valued.

Where this lives today

The branch cut is a working tool across modern physics. In quantum field theory and many-body physics a propagator $G(E)$ is analytic in the complex-energy plane except for a cut along the real axis marking the continuum of multi-particle states; the discontinuity $\Delta G(E)=G(E+i0)-G(E-i0)=2i\,\mathrm{Im}\,G(E+i0)$ is the spectral density, fixed by unitarity through the optical theorem, and Cauchy's theorem wrapped around the cut yields the Kramers–Kronig and dispersion relations that reconstruct a scattering amplitude's real part from measured cross sections. Lattice-QCD and condensed-matter groups labour over the analytic continuation of numerical Euclidean correlators back across such cuts to extract spectral functions — an ill-posed problem precisely because the cut structure must be respected. In aerodynamics the Joukowski and $\sqrt{z^2-1}$ maps place branch points at the edges of an aerofoil, and the Kutta condition demands the flow not be singular at one of them. Elliptic and hyperelliptic surfaces are the native habitat of integrable systems and Seiberg–Witten theory, where the genus and periods encode the low-energy physics — and the same $\sqrt z$ that fails at $z=0$ sets the threshold behaviour of every two-body cross section.

The physical picture

Walk once around the origin while continuously reading off $\arg z$ on a protractor. You return to the same spatial point, but the protractor now reads $\theta+2\pi$ . Any function that cares about $\theta$ rather than only $\theta\bmod 2\pi$ — the half-angle $\theta/2$ inside $\sqrt z$ , or $\theta$ itself inside $\log z$ — comes back changed. That memory of the loop is monodromy, and the point you encircled is the branch point. The first figure makes the sign flip concrete: at $z=1$ the square root starts at $+1$ and returns as $-1$ , so it takes two loops to get home.

Figure

Monodromy of √z: encircling the branch point z = 0 once sends θ → θ + 2π, so √z = √r·e^(iθ/2) gains the factor e^(iπ) = −1 and the value at z = 1 flips from +1 to −1; a second loop restores it, which is exactly why the square root needs two sheets

The branch cut is the bookkeeping that makes a single value stick. Think of it as the slit edge of a sheet of paper you have cut and re-glued: nothing intrinsic happens at a generic point of the cut for the surface, but if you insist on living in one flat plane the slit becomes a real discontinuity. Slide the cut and the discontinuity slides with it — proof that the cut is convention, not geometry, while the endpoints it must join are the immovable branch points. To hunt for a branch point at $z_0$ you encircle it and test whether the function returns; to test infinity you substitute $z=1/w$ and examine $w=0$ , which is how $\log z$ and $\sqrt z$ turn out to branch at $\infty$ even while looking tame for finite $z$ . The second figure is the payoff of this whole machine: hugging a cut turns the jump across it into the answer to a real integral.

Figure

Keyhole contour hugging the branch cut of the factor z^(a−1) on the positive real axis: the top edge (arg z = 0) and the returning bottom edge (arg z = 2π) differ by the phase e^(2πi(a−1)), and the enclosed pole at z = −1 evaluates the real integral to π/sin(πa), which equals π at a = ½

The core equations

The whole subject starts from one blunt fact: the argument of a complex number is only defined modulo $2\pi$ , so the logarithm — the inverse of the exponential — cannot be a function in the single-valued sense. It is a countable family of values, one per winding.

\log z = \ln|z| + i\,\arg z, \qquad \arg z = \theta + 2\pi k,\ \ k\in\mathbb{Z} \qquad (1)

where: $|z|$ — the modulus $r=\sqrt{x^2+y^2}$ , single-valued and positive; $\arg z$ — the multivalued argument, defined only mod $2\pi$ ; $k$ — the integer labelling which sheet or branch you are on; $\ln$ — the ordinary real logarithm of a positive number. Reading it: the real part is harmless, the multivaluedness lives entirely in the angle. To get a genuine function we choose one branch by restricting the angle. The standard choice is the principal logarithm, cut along the negative real axis.

\operatorname{Log} z = \ln|z| + i\,\theta,\qquad -\pi < \theta \le \pi \qquad (2)

where: $\operatorname{Log}$ — the principal value (capital L), single-valued and analytic on $\mathbb{C}\setminus(-\infty,0]$ ; $\theta$ — the principal argument pinned to $(-\pi,\pi]$ . Reading it: analytic on the cut plane, but discontinuous by $2\pi i$ as you step across the negative real axis. For the square root the same $2\pi$ ambiguity halves into a sign. One circuit of the origin advances the half-angle by $\pi$ and flips the value — the algebraic heart of a branch point.

z^{1/2} = \sqrt r\,e^{i\theta/2},\qquad \theta\to\theta+2\pi \;\Rightarrow\; z^{1/2}\to e^{i\pi}z^{1/2} = -\,z^{1/2} \qquad (3)

where: $r=|z|$ — unaffected by the loop; $\theta/2$ — the half-angle, which advances by $\pi$ per full circuit and so flips the sign. Reading it: two loops ( $\theta\to\theta+4\pi$ ) give $e^{2\pi i}=1$ and restore the value — the surface closes after two sheets. Because a branch is discontinuous across its cut, the size of that discontinuity is a well-defined quantity. It is the material of every dispersion relation and every keyhole integral.

\Delta f(x) \equiv f(x+i0^+) - f(x-i0^-) \qquad (4)

where: $f(x\pm i0)$ — the limits of the chosen branch approaching a real point $x$ on the cut from above and below; $\Delta f$ — the jump, zero off the cut and nonzero on it. Reading it: the jump is the observable trace of the branch point living at the cut's end. The cleanest test for a logarithmic branch point is to integrate $\mathrm d\log$ around a small loop: a nonzero answer is the failure of single-valuedness.

\oint_{|z-z_0|=\varepsilon}\frac{\mathrm dz}{z-z_0} = 2\pi i \;\Rightarrow\; \operatorname{Log}(z-z_0)\ \text{gains}\ 2\pi i \qquad (5)

where: $\varepsilon$ — the radius of the small test loop; $z_0$ — the suspected branch point. Reading it: the $2\pi i$ is the winding number times $2\pi i$ ; it is why $\log$ needs infinitely many sheets, one per integer winding. Riemann's reformulation makes the multivalued map single-valued by promoting it to a curve in two complex dimensions — the graph, read as a surface.

w^n = z \quad\Longleftrightarrow\quad \{(z,w):\,w^n-z=0\}\ \subset\ \mathbb{C}^2 \qquad (6)

where: $w$ — the dependent value ( $=z^{1/n}$ ); on the surface the projection $(z,w)\mapsto w$ is an honest single-valued function; $n$ — the number of sheets, equal to the number of branches. Reading it: the multifunction becomes a function once you agree to live on the curve, not on the $z$ -plane. Finally, the topology of the compact surface is fixed by counting sheets and branch points through Riemann–Hurwitz, with the hyperelliptic family as the worked case.

\chi = n\,\chi_0 - \sum_{p}(e_p-1),\qquad y^2=\prod_{i=1}^{2g+2}(z-e_i)\ \Rightarrow\ \text{genus }g \qquad (7)

where: $\chi=2-2g$ — the Euler characteristic of the closed surface of genus $g$ ; $\chi_0=2$ — the Euler characteristic of the sphere base; $n$ — sheets; $e_p$ — the ramification index at a branch point $p$ ; $e_i$ — the distinct branch points of a hyperelliptic curve. Reading it: two sheets over two branch points give a sphere ( $g=0$ ); two sheets over four give a torus ( $g=1$ ).

Derivation ABuilding the two-sheeted Riemann surface of the square root

The goal is to turn the multivalued $\sqrt z$ into a single-valued analytic function by constructing its Riemann surface explicitly — starting from one honest branch on a slit plane, adding the second branch, and gluing the two slit copies edge-to-edge so that the branch cut stops being a discontinuity and becomes merely the seam where two sheets meet. We then identify the resulting closed surface topologically and confirm it by counting with Riemann–Hurwitz, so that the abstract phrase "double cover of the sphere" becomes something you can see.

Step 1. Fix one branch on the plane slit along the positive real axis: $w_0(z)=\sqrt r\,e^{i\theta/2}$ with $0\le\theta<2\pi$ . This $w_0$ is analytic and single-valued on the slit plane, but jumps from $\sqrt r$ just above the cut to $-\sqrt r$ just below it. Why: a single flat sheet can only hold one of the two roots continuously, and the price is a discontinuity somewhere — we have parked it on the positive real axis.

Step 2. Introduce the second branch $w_1(z)=-w_0(z)=\sqrt r\,e^{i(\theta/2+\pi)}$ . On its own slit copy it too is single-valued and analytic, and together $\{w_0,w_1\}$ exhaust both square roots at every $z\neq0$ . Why: the two branches differ only by the sign $e^{i\pi}=-1$ picked up in one loop, so listing both is the same as tracking the loop that swaps them.

Step 3. Match the jumps. Just above sheet 0's cut, $w_0\to\sqrt r$ ; just below sheet 0's cut, $w_0\to-\sqrt r$ , which is precisely the value of $w_1$ just above its cut. So glue the upper lip of sheet 0's slit to the lower lip of sheet 1's, and vice versa. Why: gluing edges where the values already agree makes the join continuous — analytic continuation across the seam is automatic once the boundary values coincide.

Step 4. Read off the consequence of the gluing: crossing the cut once carries you smoothly from sheet 0 to sheet 1, and crossing again returns you to sheet 0. The cut has ceased to be a discontinuity; it is now the internal seam of a connected two-sheeted surface. Why: this is the whole point of the construction — the "curtain" is gone, and $w$ is single-valued and analytic on the glued object.

Step 5. Identify the surface. Both $z=0$ and $z=\infty$ are order-two branch points where the two sheets join. Riemann–Hurwitz with $n=2$ sheets over the sphere and two simple branch points gives $\chi=2\cdot2-2=2$ , hence $g=0$ : topologically a sphere.

\mathcal R_{\sqrt z}=\{(z,w):w^2=z\}\cong S^2,\qquad w=\sqrt z\ \text{single-valued on }\mathcal R

Two sheets glued along one cut with two simple branch points give $\chi=2$ , genus $0$ ; this is consistent with the parametrization $z=w^2$ , which maps the $w$ -sphere bijectively onto $\mathcal R$ .

Go deeper · Derivation B · 1 minEvaluating a real integral by hugging a branch cut

Derivation BEvaluating a real integral by hugging a branch cut

The goal is to compute the real integral $\int_0^\infty x^{a-1}/(1+x)\,\mathrm dx$ for $0<a<1$ using the very multivaluedness that seems to obstruct it: give $z^{a-1}$ a branch cut on the positive real axis, run a keyhole along both lips, and the mismatched phase between the top and bottom edges — the jump across the cut — is exactly what fails to cancel and leaves the wanted integral, which residues then pin down. It is the workhorse technique behind a whole family of integrals in mathematical physics.

Step 1. Choose the branch $z^{a-1}=e^{(a-1)\operatorname{Log}z}$ with $\arg z\in(0,2\pi)$ and cut on the positive real axis, and take the keyhole: out along the top lip, around a big circle $R\to\infty$ , back along the bottom lip, and around a small circle $\varepsilon\to0$ at the origin, enclosing the single pole at $z=-1$ . Because $0<a<1$ , both circular arcs contribute nothing in the limit. Why: the exponents $x^{a}$ and $x^{a-1}$ kill the large and small arcs respectively, so only the two straight edges survive.

Step 2. On the top edge $\arg z=0$ , so $z^{a-1}=x^{a-1}$ and the contribution is $\int_0^\infty x^{a-1}/(1+x)\,\mathrm dx\equiv I$ . On the bottom edge $\arg z=2\pi$ , so $z^{a-1}=x^{a-1}e^{2\pi i(a-1)}$ , and running inward reverses the direction, giving $-e^{2\pi i(a-1)}I$ . Why: the two edges are the same real ray but on opposite lips of the cut, so they carry the same $|z^{a-1}|$ but differ by the full monodromy phase $e^{2\pi i(a-1)}$ — the jump made visible.

Step 3. Evaluate the enclosed residue. On this branch $-1=e^{i\pi}$ , so $\operatorname{Res}_{z=-1}\dfrac{z^{a-1}}{1+z}=(-1)^{a-1}=e^{i\pi(a-1)}$ , and the contour integral equals $2\pi i\,e^{i\pi(a-1)}$ . Why: the pole sits at the midpoint angle $\pi$ between the two lips, so its phase is exactly halfway between the top and bottom edge phases — a useful internal consistency check.

Step 4. Assemble. The two edges sum to $(1-e^{2\pi i(a-1)})\,I$ and equal the residue term:

\big(1-e^{2\pi i(a-1)}\big)\,I = 2\pi i\,e^{i\pi(a-1)}.

Factor $e^{i\pi(a-1)}$ out of the bracket, $1-e^{2\pi i(a-1)}=-e^{i\pi(a-1)}\big(e^{i\pi(a-1)}-e^{-i\pi(a-1)}\big)=-2i\,e^{i\pi(a-1)}\sin\pi(a-1)$ , and cancel:

I=\frac{2\pi i}{-2i\sin\pi(a-1)}=\frac{-\pi}{\sin\pi(a-1)}=\frac{\pi}{\sin\pi a},\qquad 0<a<1.

As a check: numerically $I(\tfrac12)=\pi=3.14159$ and $I(\tfrac13)=2\pi/\sqrt3=3.62760$ ; analytically the result matches the reflection formula $\Gamma(a)\Gamma(1-a)=\pi/\sin\pi a$ , since $I=B(a,1-a)$ .

Go deeper · Derivation C · 2 minThe monodromy test and why a dogbone cut suffices

Derivation CThe monodromy test and why a dogbone cut suffices

The goal is to make branch-point detection mechanical and to explain the economy of the finite "dogbone" cut. We read off the universal winding factor $e^{2\pi i\alpha}$ that any power $z^\alpha$ picks up around the origin, recover the logarithm's $2\pi i$ shift as its limit, and apply the same loop-counting to $f(z)=\sqrt{(z-a)(z-b)}$ to show why one finite segment joining its two branch points restores single-valuedness everywhere else — though each branch point individually is a genuine obstruction.

Step 1. For $z^\alpha=r^\alpha e^{i\alpha\theta}$ , one counterclockwise loop sends $\theta\to\theta+2\pi$ and multiplies the value by $e^{2\pi i\alpha}$ . This equals $1$ exactly when $\alpha\in\mathbb Z$ (no branch point), equals $-1$ for $\alpha=\tfrac12$ , and closes after $n$ loops for $\alpha=1/n$ . Why: the winding factor is the single number that decides both whether $z=0$ is a branch point and how many sheets the surface needs — its order is the smallest $n$ with $e^{2\pi in\alpha}=1$ .

Step 2. Take the logarithmic limit. Writing $z^\alpha=e^{\alpha\log z}$ and expanding for small $\alpha$ , the per-loop change $e^{2\pi i\alpha}-1\approx 2\pi i\alpha$ divided by $\alpha$ tends to $2\pi i$ , which is exactly the shift $\log z\to\log z+2\pi i$ from equation (5). Why: the transcendental (logarithmic) branch point is the $\alpha\to0$ shadow of the algebraic ones — never closing, hence infinitely many sheets.

Step 3. Now $f(z)=\sqrt{(z-a)(z-b)}=\exp\!\big[\tfrac12(\operatorname{Log}(z-a)+\operatorname{Log}(z-b))\big]$ . A loop enclosing only $a$ sends the $(z-a)$ factor's angle up by $2\pi$ , contributing $e^{i\pi}=-1$ , so $f\to-f$ : single-valuedness fails and a cut must emanate from $a$ . Likewise for $b$ . Why: each factor is an independent order-two branch point, so a loop around either one alone is fatal.

Step 4. A loop enclosing both $a$ and $b$ raises both angles by $2\pi$ , contributing $(-1)(-1)=+1$ : $f$ returns to itself. Therefore $f$ is single-valued on the plane minus the segment $[a,b]$ , and a finite cut along that segment — the "dogbone" — suffices; equivalently one may run two cuts to infinity. Why: forbidding only the loops that enclose exactly one branch point is enough, and the shortest such barrier is the segment joining the pair. The associated surface is again a sphere ( $g=0$ ), while four branch points would force a torus.

Limiting cases

The order of a branch point fixes both the per-loop factor and the number of sheets the Riemann surface needs.

Case	Result	Meaning
Square root	$\sqrt z\ \xrightarrow{\ \theta\to\theta+2\pi\ }\ -\sqrt z$	Order-2 (algebraic) branch points at $0$ and $\infty$ ; two sheets; two loops restore the value; surface is a sphere.
$n$ -th root	$z^{1/n}\ \xrightarrow{\ \theta\to\theta+2\pi\ }\ e^{2\pi i/n}z^{1/n}$	Order- $n$ branch points; $n$ sheets; compact surface closing after $n$ circuits.
Logarithm	$\log z\ \xrightarrow{\ \theta\to\theta+2\pi\ }\ \log z+2\pi i$	Logarithmic (transcendental) branch points at $0$ and $\infty$ ; infinitely many sheets; the helicoid never closes.
Irrational power	$z^{\alpha},\ \alpha\notin\mathbb Q\ \xrightarrow{\ \theta\to\theta+2\pi\ }\ e^{2\pi i\alpha}z^{\alpha}$	Every loop lands on a new value; infinitely many sheets, like the logarithm's.

Common trapThe branch cut is a real, fixed feature of the function

It is not — and the whole subject turns on the distinction. The branch points are intrinsic: they are exactly where monodromy is nontrivial, and you cannot move them (for $\log z$ they are $0$ and $\infty$ ). The branch cut joining them is your choice of curtain and may be slid, bent, or replaced freely: two different cuts describe the same function and build the same surface, and the discontinuity moves with the cut, proving it is convention rather than geometry. A second, related trap: a branch point is not a pole. A pole is an isolated singularity with a single-valued Laurent expansion on a punctured disc; a branch point has no single-valued punctured neighbourhood at all, so no Laurent series exists. Indeed $\sqrt z$ is finite and continuous at $z=0$ , yet $0$ is still a branch point — a different species of singularity entirely.

Worked examples

1 · A keyhole integral to a real number (confidence builder)

Evaluate $\displaystyle\int_0^\infty\frac{\mathrm dx}{\sqrt x\,(1+x)}$ and then $\displaystyle\int_0^\infty\frac{x^{-2/3}}{1+x}\,\mathrm dx$ , using the master result from the derivation.

Step 1. Recognize the family. Both integrands have the shape $x^{a-1}/(1+x)$ . The first has $x^{-1/2}=x^{a-1}$ , so $a-1=-\tfrac12$ , i.e. $a=\tfrac12$ ; the second has $x^{-2/3}$ , so $a=\tfrac13$ . Both satisfy $0<a<1$ , so the keyhole result applies.

Step 2. Apply $\displaystyle\int_0^\infty\frac{x^{a-1}}{1+x}\,\mathrm dx=\frac{\pi}{\sin\pi a}$ . For $a=\tfrac12$ : $\sin(\pi/2)=1$ , so the integral is $\pi/1=\boldsymbol{\pi\approx3.14159}$ . For $a=\tfrac13$ : $\sin(\pi/3)=\sqrt3/2$ , so the integral is $\pi/(\sqrt3/2)=2\pi/\sqrt3=\boldsymbol{3.62760}$ .

Step 3. Sanity-check the first independently. Substitute $x=u^2$ : the integral becomes $2\int_0^\infty\mathrm du/(1+u^2)=2\cdot\tfrac{\pi}{2}=\pi$ , so the elementary route and the branch-cut machinery agree exactly. The two integrals are π and 2π/√3, and the cut's phase mismatch was the essential ingredient — without it the top and bottom edges would cancel to zero.

2 · Reading the genus off the branch points

How many handles does the Riemann surface of $w=\sqrt{P(z)}$ have when $P$ is a cubic or a quartic with distinct roots? Compare with $\sqrt z$ .

Step 1. Count sheets and branch points. A square root is a two-sheeted cover ( $n=2$ ), and its branch points are the simple zeros of $P$ plus possibly $z=\infty$ . For a quartic the behaviour at infinity is $w\sim z^2$ , single-valued, so $\infty$ is not a branch point: $B=4$ . For a cubic, $w\sim z^{3/2}$ at infinity is multivalued, so $\infty$ counts too: $B=3+1=4$ again.

Step 2. Apply Riemann–Hurwitz, $\chi=n\chi_0-\sum_p(e_p-1)=2\cdot2-B\cdot1=4-B$ . With $B=4$ in both cases, $\chi=4-4=\boldsymbol{0}$ , and $\chi=2-2g$ gives $g=1$ : a torus. This is precisely why elliptic integrals such as $\int\mathrm dx/\sqrt{(1-x^2)(1-k^2x^2)}$ — a quartic under the root — live naturally on a genus-1 surface, and why their inverses are doubly periodic.

Step 3. Contrast with $\sqrt z$ : here $P(z)=z$ has one finite branch point plus one at infinity, $B=2$ , so $\chi=4-2=2$ and $g=0$ — a sphere, matching Derivation 1. The pattern: two sheets over $B$ simple branch points give genus $g=(B-2)/2$ , so $2\to$ sphere, $4\to$ torus, $6\to$ genus-2, exactly the hyperelliptic ladder.

Physics of the portrait

Branch points were handled piecemeal by Euler, Cauchy, and Puiseux, but it took Riemann to see that the right object was not the cut plane at all — it was a new surface on which the multivalued function became single-valued, and whose shape encoded the function's deepest global properties.

Origins · 1 minBernhard Riemann and how this was discovered

Bernhard Riemann

1826–1866 · complex analysis and the geometry of multivalued functions

Georg Friedrich Bernhard Riemann, born in 1826 in the Kingdom of Hanover, reshaped analysis, geometry, and number theory in a career of barely fifteen years before tuberculosis killed him at 39. His 1851 doctoral dissertation, written under Gauss at Göttingen, introduced what we now call Riemann surfaces: rather than fight the multivaluedness of $\sqrt z$ or $\log z$ with cuts, he built a many-sheeted covering of the plane on which the function is honestly single-valued and analytic — trading an algebraic nuisance for a geometric object. The 1857 memoir on Abelian functions turned this into a theory: he classified such surfaces by their connectivity, introduced the genus as the count of independent cycles, and tied a function's analysis to its surface's topology — the first deep marriage of analysis and topology. The same thesis gave the Cauchy–Riemann equations their modern role and the mapping theorem that bears his name; his 1859 paper launched the zeta function and the still-unsolved hypothesis, and his 1854 lecture supplied the curved-space framework Einstein would need for general relativity. That branch cuts, an X-ray dispersion relation, and the spacetime of gravitation all trace to one short-lived Göttingen lecturer measures how far a single geometric idea can reach.

History

1748

— Euler wrestles with the infinitely many values of the logarithm of a complex number

1825

— Cauchy's memoir on integrals between imaginary limits exposes path dependence

1850

— Puiseux analyzes the branch points of algebraic functions by tracking loops

1851

— Riemann's dissertation introduces the multi-sheeted surfaces that bear his name

1857

— Riemann's theory of Abelian functions ties genus to the topology of the surface

1891

— Hurwitz proves the ramification formula relating genus to branch points

Going further

Texts. Stone & Goldbart, Mathematics for Physics, ch. on functions of a complex variable (branch points, cuts, Riemann surfaces) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., ch. 11 · Hassani, Mathematical Physics, chs. on complex analysis · Nakahara, Geometry, Topology and Physics, ch. 8 (Riemann surfaces and genus) · Byron & Fuller, Mathematics of Classical and Quantum Physics · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, 3rd ed., ch. 24 · Ablowitz & Fokas, Complex Variables · Needham, Visual Complex Analysis (geometric treatment of multifunctions). Key results to keep in mind: the winding factor for $z^\alpha$ is $e^{2\pi i\alpha}$ , and the $2\pi i$ shift of $\log z$ per loop follows from $\oint\mathrm dz/z=2\pi i$ ; the jump of $\sqrt z$ across the negative real axis is $\Delta\sqrt z=2i\sqrt{|x|}$ ; the keyhole formula gives $\int_0^\infty x^{a-1}/(1+x)\,\mathrm dx=\pi/\sin\pi a$ , with numerical values $\pi$ at $a=\tfrac12$ and $2\pi/\sqrt3=3.62760$ at $a=\tfrac13$ , consistent with the reflection formula $\Gamma(a)\Gamma(1-a)=\pi/\sin\pi a$ ; and Riemann–Hurwitz $\chi=2\cdot2-B$ gives genus $0$ for $\sqrt z$ ( $B=2$ ) and genus $1$ for $\sqrt{P}$ with a quartic or cubic $P$ ( $B=4$ ).

Prerequisites. Complex Numbers and the Complex Plane · Elementary Complex Functions · Functions of a Complex Variable and Limits · Analytic Functions and the Cauchy–Riemann Equations · Complex Integration and Contour Integrals. Leads to. Laurent Series · Zeros, Poles, and Isolated Singularities · Complex Integration and Contour Integrals.

Ask the tutor: Ask me to walk a loop around any branch point you choose — $\sqrt{z^2-1}$ , $\log(z-i)$ , or an elliptic integrand — and watch the value change, the cut appear, and the surface assemble itself.

What is the difference between a branch point and a branch cut, in one line?

The branch point is the disease; the branch cut is one of many possible bandages. A branch point is an intrinsic location — fixed by where a loop changes the function's value — and it cannot be moved: for $\sqrt z$ it is $z=0$ and $z=\infty$ . A branch cut is a curve you draw joining branch points to forbid the loops that would expose the multivaluedness, and it is entirely your choice: straight or curved, finite or running to infinity. Change the cut and you change where the discontinuity sits, but never the function itself and never its Riemann surface. If you remember only one thing, remember that points are physics and cuts are bookkeeping.

Why is the value of the discontinuity across a cut physically important?

Because in physics the cut of a response function is the spectrum of real states, and the jump across it is their density. For a propagator $G(E)$ analytic off the real axis, the discontinuity $\Delta G=G(E+i0)-G(E-i0)=2i\,\mathrm{Im}\,G$ is the spectral function, tied by unitarity (the optical theorem) to decay rates and scattering cross sections. Cauchy's theorem applied to a contour hugging the cut then converts that jump into a dispersion (Kramers–Kronig) relation, expressing the real part of an amplitude as an integral of its imaginary part over the cut. So the "defect" of a cut is exactly the channel through which measured spectra enter analytic theory — the geometry of the cut is the analytic encoding of what particles the theory contains.

How do I actually find where a function's branch points are?

Test each candidate with a small loop. Encircle a finite point $z_0$ by $|z-z_0|=\varepsilon$ and continue the function once around: if it fails to return to its starting value, $z_0$ is a branch point, and the factor it picks up tells you its order. For infinity, substitute $z=1/w$ and examine $w=0$ the same way — this is how $\sqrt z$ and $\log z$ reveal branch points at $\infty$ despite looking tame for large finite $z$ . For a product like $\sqrt{(z-a)(z-b)}$ , test each factor separately and then test loops enclosing several at once, because the combined winding can cancel. The rule underneath is always monodromy: a branch point is any point some loop cannot forgive.

Why does log z need infinitely many sheets while z^(1/n) needs only n?

Because the per-loop factor is different in kind. Around the origin $z^{1/n}$ is multiplied by $e^{2\pi i/n}$ , and after $n$ loops that factor becomes $e^{2\pi i}=1$ : you are home, so the surface closes into a compact $n$ -sheeted cover. But $\log z$ is not multiplied — it is shifted by $2\pi i$ every loop, and no number of $2\pi i$ shifts ever returns to zero. So the sheets march off forever, one per integer winding number, and the surface is an infinite helicoid, a spiral staircase over the punctured plane that never reconnects. Algebraic branch points close; the logarithmic (transcendental) branch point does not.

Is the genus of a Riemann surface something I can compute by hand?

Yes, by counting. For a square root $w=\sqrt{P(z)}$ (two sheets), list the simple branch points: the distinct roots of $P$ , plus $z=\infty$ when the degree of $P$ is odd. Call the total $B$ . Riemann–Hurwitz gives the Euler characteristic $\chi=2\cdot2-B$ , and $\chi=2-2g$ then fixes the genus $g=(B-2)/2$ . So two branch points give a sphere ( $g=0$ ), four a torus ( $g=1$ ), six genus 2, up the hyperelliptic ladder. For an $n$ -th root, replace the "2 sheets" by $n$ and weight each branch point by its ramification $e_p-1$ . The arithmetic is elementary; the payoff — knowing the surface's shape — is not.

If √z is finite at z = 0, how can 0 be a singularity at all?

Because "singularity" here means a failure of analyticity in the single-valued sense, not a blow-up. At $z=0$ the value $\sqrt 0=0$ is perfectly finite and even continuous — but there is no punctured disc around $0$ on which $\sqrt z$ is single-valued, so it has no Taylor or Laurent expansion there and no well-defined derivative in the ordinary sense. That failure of single-valuedness is the singularity. It is a genuinely different species from a pole (which blows up but stays single-valued) or an essential singularity (single-valued with an infinite Laurent tail). Branch points are the third kind, and they are why the whole apparatus of cuts and surfaces exists.