Analytic Continuation and Poles of the Gamma Function

The factorial $n!$ counts arrangements of $n$ objects and is defined only on the non-negative integers; the gamma function is its smooth, analytic interpolation, pinned by $\Gamma(n+1)=n!$ . Euler's integral $\Gamma(z)=\int_0^\infty t^{z-1}e^{-t}\,\mathrm{d}t$ turns that interpolation into a concrete analytic function — but only on the right half-plane $\operatorname{Re}z>0$ , because the factor $t^{z-1}$ becomes non-integrable at $t=0$ once $\operatorname{Re}z\le0$ . The central problem of this article is how to push $\Gamma$ past that wall in the one analytically faithful way the plane allows, and to read off the singular structure that results. The engine of continuation is the functional equation $\Gamma(z+1)=z\,\Gamma(z)$ , a single integration by parts. Rewritten as $\Gamma(z)=\Gamma(z+1)/z$ , it re-expresses $\Gamma$ on the strip $-1<\operatorname{Re}z\le0$ using values where the integral already converges; iterate, and the function marches leftward by any integer. Each backward step divides by a factor that vanishes at an integer, and that division is precisely what stamps a simple pole onto the otherwise smooth landscape. By the identity theorem the result is unique — exactly one meromorphic function on $\mathbb{C}$ agrees with Euler's integral on the half-plane, so "the" continuation is well posed. Two symmetries then organize everything. The reflection formula $\Gamma(z)\Gamma(1-z)=\pi/\sin\pi z$ places the poles exactly where $\sin\pi z$ vanishes and proves $\Gamma$ has no zeros at all, so $1/\Gamma$ is entire. The Legendre duplication formula, and its parent the Gauss multiplication theorem, tie together gamma values at rationally spaced arguments. With the Weierstrass product and the constant $\gamma$ , these make $\Gamma$ the prototype every special function in mathematical physics is built to imitate.

Key takeaways

Euler's integral $\Gamma(z)=\int_0^\infty t^{z-1}e^{-t}\,\mathrm{d}t$ is analytic only for $\operatorname{Re}z>0$ ; the recurrence $\Gamma(z+1)=z\Gamma(z)$ extends it to a meromorphic function on all of $\mathbb{C}$ .
The continuation has simple poles at $z=0,-1,-2,\dots$ with residue $(-1)^n/n!$ at $z=-n$ , and it is nowhere zero — so $1/\Gamma(z)$ is entire.
The reflection formula $\Gamma(z)\Gamma(1-z)=\pi/\sin(\pi z)$ ties $z$ to $1-z$ , encodes every pole, and delivers $\Gamma(\tfrac12)=\sqrt{\pi}$ .
The Legendre duplication formula $\Gamma(z)\Gamma(z+\tfrac12)=2^{1-2z}\sqrt{\pi}\,\Gamma(2z)$ is the $m=2$ case of the Gauss multiplication theorem and supplies all half-integer values.
The Weierstrass product $1/\Gamma(z)=z\,e^{\gamma z}\prod_n(1+z/n)e^{-z/n}$ exhibits $1/\Gamma$ as the canonical entire function over the negative integers and brings in $\gamma\approx0.5772$ .
These properties underpin half-integer factorials, $n$ -ball volumes, Bessel normalizations, and the $1/\varepsilon$ poles of dimensional regularization.
Key idea: the recurrence $\Gamma(z+1)=z\Gamma(z)$ is at once the tool that continues $\Gamma$ across the whole plane and the source of its poles — dividing by $z,z+1,\dots$ manufactures simple poles at $z=0,-1,-2,\dots$ with residues $(-1)^n/n!$ .

The core insight

The recurrence $\Gamma(z+1)=z\Gamma(z)$ is simultaneously the tool that continues $\Gamma$ to the whole plane and the source of its poles: dividing by $z,z+1,\dots$ produces simple poles at $z=0,-1,-2,\dots$ with residues $(-1)^n/n!$ .

Where this lives today

The analytic continuation of $\Gamma$ is running inside modern calculations right now. In quantum field theory, dimensional regularization computes loop integrals in $d=4-2\varepsilon$ dimensions, and every ultraviolet divergence surfaces as a pole of a gamma function in $\varepsilon$ ; the Laurent coefficients $-\gamma$ and $\pi^2/12$ throughout the two-loop QCD literature are literally the expansion of $\Gamma(\varepsilon)$ at its simplest pole. In the theory of Bessel and spherical Bessel functions, the entireness of $1/\Gamma$ makes the power-series coefficients $1/\Gamma(k+\nu+1)$ finite for every order $\nu$ , and its zeros at negative integers enforce $J_{-m}=(-1)^m J_m$ . Half-integer gamma values from the duplication formula set the normalizations of associated Legendre functions and the small-argument $j_\ell(x)\sim x^\ell/(2\ell+1)!!$ . Further afield, the same continuation underlies the functional equation of the Riemann zeta function — $\zeta(s)$ and $\zeta(1-s)$ tied by exactly the reflection-formula machinery — and the analytic regularization of divergent sums in string amplitudes and Casimir-energy calculations. Whenever a physicist writes a divergent quantity as the value of a meromorphic function and reads off a finite residue, they are standing on the continuation built here.

The physical picture

Picture $\Gamma$ as a function known cleanly on the right half-plane and then pulled leftward by the rule $\Gamma(z)=\Gamma(z+1)/z$ . On $\operatorname{Re}z>0$ the Euler integral gives a smooth, positive curve on the real axis, dipping to a single minimum $\Gamma(1.4616)\approx0.8856$ between the twin values $\Gamma(1)=\Gamma(2)=1$ and climbing to the factorials beyond. Each time the continuation crosses an integer going left, it divides by a factor that vanishes there — and that lone division is what turns the smooth curve into a spike. The first figure is the real-axis portrait: a gentle bowl on the right, and a run of simple poles at $z=0,-1,-2,-3$ whose residues alternate in sign and shrink factorially.

Figure

The real gamma function Γ(x): smooth and positive for x greater than 0 with a single minimum Γ(1.4616) ≈ 0.8856, and simple poles at x = 0, -1, -2, -3 whose residues +1, -1, +1/2, -1/6 alternate in sign and shrink factorially

The reflection formula turns the picture into a mirror. Across the vertical line $\operatorname{Re}z=\tfrac12$ , the value of $\Gamma$ on one side fixes it on the other through the factor $\pi/\sin\pi z$ . The zeros of $\sin\pi z$ — one at every integer — account exactly for the poles of $\Gamma$ , while the fact that $\Gamma(1-z)$ is itself never zero forces $\Gamma$ to have no zeros anywhere. It is $1/\Gamma$ , not $\Gamma$ , that is the well-behaved entire function, with simple zeros at the non-positive integers. The second figure moves into the complex plane and shows the full analytic anatomy: a shaded half-plane where Euler's integral lives, the reflection axis, and the ladder of poles carrying the residues that make dimensional regularization tick.

Figure

The analytic structure in the complex z-plane: Euler's integral converges only for Re z greater than 0 (shaded), continuation fills the rest, and simple poles sit at z = -n on the real axis with residue Res = (-1)ⁿ/n! — decaying factorially, so 1/Γ is entire and Γ has no zeros

The core equations

The story starts with Euler's integral, the concrete analytic function that fixes what "the" gamma function is on the half-plane where it converges. Everything downstream is either an extension of this object or a computation with it.

\Gamma(z)=\int_0^\infty t^{z-1}e^{-t}\,\mathrm{d}t,\qquad \operatorname{Re}z>0 \qquad (1)

where: $z$ — complex argument, restricted to $\operatorname{Re}z>0$ ; $t$ — real integration variable. The tail at $t\to\infty$ converges for every $z$ because $e^{-t}$ crushes it; the endpoint $t\to0$ behaves like $t^{\operatorname{Re}z-1}$ and is integrable only when $\operatorname{Re}z>0$ — the wall the continuation must climb over. A single integration by parts on $(1)$ yields the functional equation — the load-bearing relation of the subject: read forward it reproduces the factorial, read backward it is the continuation machine.

\Gamma(z+1)=z\,\Gamma(z) \qquad (2)

where: $z$ — any point where $\Gamma$ is already defined; the relation propagates that definition to neighbouring strips. With $\Gamma(1)=1$ it gives $\Gamma(n+1)=n!$ by induction, so $\Gamma$ genuinely interpolates the factorial. Iterating $(2)$ backwards $n+1$ times expresses $\Gamma$ on ever-wider strips with a convergent Euler integral on top and an explicit polynomial below. This one formula both defines the continuation and displays its poles.

\Gamma(z)=\frac{\Gamma(z+n+1)}{z(z+1)\cdots(z+n)},\qquad n=0,1,2,\dots \qquad (3)

where: $n$ — a non-negative integer chosen large enough that $\operatorname{Re}(z+n+1)>0$ , so the numerator is the genuine integral. The denominator vanishes simply at $z=0,-1,\dots,-n$ ; since the numerator is nonzero there, each is a simple pole. Taking the residue of $(3)$ at each pole gives the alternating, factorially shrinking sequence that governs the local behaviour of $\Gamma$ — and, in physics, the strength of the ultraviolet divergences it encodes.

\operatorname{Res}_{z=-n}\Gamma(z)=\frac{(-1)^n}{n!},\qquad n=0,1,2,\dots \qquad (4)

where: $-n$ — location of the $n$ th simple pole. The residues $+1,-1,\tfrac12,-\tfrac16,\dots$ decay fast enough that the principal parts sum into the entire function $1/\Gamma$ ; near a pole, $\Gamma(z)\approx\dfrac{(-1)^n/n!}{z+n}$ . The reflection formula is the deepest symmetry of $\Gamma$ : it relates $z$ to $1-z$ , packages every pole into the zeros of $\sin\pi z$ , and — because its right side never vanishes — proves $\Gamma$ has no zeros at all.

\Gamma(z)\,\Gamma(1-z)=\frac{\pi}{\sin(\pi z)} \qquad (5)

where: $z$ — any non-integer complex number. Setting $z=\tfrac12$ gives $\Gamma(\tfrac12)^2=\pi$ , hence $\Gamma(\tfrac12)=\sqrt{\pi}$ ; on the critical line, $|\Gamma(\tfrac12+iy)|^2=\pi/\cosh\pi y$ . The Weierstrass product turns " $\Gamma$ has no zeros" into a construction: $1/\Gamma$ is the canonical entire function built from its simple zeros at the non-positive integers, with $\gamma$ appearing as the convergence-enforcing factor.

\frac{1}{\Gamma(z)}=z\,e^{\gamma z}\prod_{n=1}^{\infty}\left(1+\frac{z}{n}\right)e^{-z/n},\qquad \gamma\approx0.5772156649 \qquad (6)

where: $\gamma$ — Euler–Mascheroni constant, equal to $-\Gamma'(1)$ ; $n$ — index over positive integers. The factors $e^{-z/n}$ make the product converge, and they are why $\gamma$ is inseparable from the analytic structure of $\Gamma$ . The Legendre duplication formula, and its parent the Gauss multiplication theorem, connect gamma values at rationally spaced arguments — the identities that generate every half-integer factorial in physics.

\Gamma(z)\,\Gamma\!\left(z+\tfrac12\right)=2^{1-2z}\sqrt{\pi}\;\Gamma(2z),\qquad \Gamma\!\left(n+\tfrac12\right)=\frac{(2n)!}{4^n\,n!}\sqrt{\pi} \qquad (7)

where: $z$ — any complex number keeping both sides finite; $n$ — non-negative integer. Duplication is the $m=2$ case of the Gauss multiplication theorem $\prod_{k=0}^{m-1}\Gamma(z+\tfrac{k}{m})=(2\pi)^{(m-1)/2}m^{1/2-mz}\Gamma(mz)$ .

Derivation AIntegration by parts and the leftward march to the poles

The goal is to build the entire continuation from one calculus move and then read off its singularities. We integrate Euler's integral by parts to get $\Gamma(z+1)=z\Gamma(z)$ , rearrange it into a machine that shifts the domain leftward one unit at a time, iterate to any strip $\operatorname{Re}z>-(n+1)$ , and take a residue to prove the poles at the non-positive integers are simple with residue $(-1)^n/n!$ . The point to keep in view is that a single relation does two jobs at once — it defines $\Gamma$ where the integral has died, and the very act of division is what creates the poles.

Step 1. Start from $\Gamma(z+1)=\int_0^\infty t^{z}e^{-t}\,\mathrm{d}t$ and integrate by parts with $u=t^z$ , $\mathrm{d}v=e^{-t}\,\mathrm{d}t$ , giving $\Gamma(z+1)=\big[-t^z e^{-t}\big]_0^\infty+z\int_0^\infty t^{z-1}e^{-t}\,\mathrm{d}t$ . The boundary term vanishes at both ends for $\operatorname{Re}z>0$ , leaving $\Gamma(z+1)=z\,\Gamma(z)$ . Why: this is the only input; the boundary term dies because $t^z e^{-t}\to0$ at infinity and $t^{\operatorname{Re}z}\to0$ at the origin when $\operatorname{Re}z>0$ .

Step 2. Rearrange to $\Gamma(z)=\Gamma(z+1)/z$ . On the strip $-1<\operatorname{Re}z\le0$ the right side is well defined — its numerator $\Gamma(z+1)$ sits where the integral converges — so this equation defines $\Gamma$ one unit to the left of where we started. Why: continuation is not a new axiom but a rearrangement; we are forced into it because analyticity leaves no other consistent choice.

Step 3. Iterate: $\Gamma(z)=\Gamma(z+1)/z=\Gamma(z+2)/[z(z+1)]=\cdots=\Gamma(z+n+1)/[z(z+1)\cdots(z+n)]$ . Choosing $n$ so that $\operatorname{Re}(z+n+1)>0$ makes the numerator an honest convergent integral, valid for all $z$ with $\operatorname{Re}z>-(n+1)$ except where the denominator vanishes. Why: each factor extends the domain by one unit, so any point in the plane is reached with finite $n$ — the continuation covers all of $\mathbb{C}$ .

Step 4. Near $z=-n$ multiply by $(z+n)$ and take the limit: $\operatorname{Res}_{z=-n}\Gamma=\lim_{z\to-n}(z+n)\,\Gamma(z+n+1)/[z(z+1)\cdots(z+n)]=\Gamma(1)/[(-n)(-n+1)\cdots(-1)]$ . The surviving product of $n$ negative factors is $(-1)^n n!$ and $\Gamma(1)=1$ , so $\operatorname{Res}_{z=-n}\Gamma=(-1)^n/n!$ . Why: the $(z+n)$ cancels exactly one denominator factor, confirming a simple pole, and the leftover numbers give the alternating factorial residues. *As a check: the residue at $z=0$ is $1$ , and near $z=0$ the Laurent series gives $\Gamma(z)=\Gamma(z+1)/z\approx1/z$ ; the computed values $\Gamma(-\tfrac12)=-2\sqrt{\pi}$ and $\Gamma(-\tfrac32)=\tfrac43\sqrt{\pi}$ both obey the recurrence.*

Go deeper · Derivation B · 1 minThe reflection formula from the beta integral

Derivation BThe reflection formula from the beta integral

The goal is to prove $\Gamma(z)\Gamma(1-z)=\pi/\sin\pi z$ , the identity that simultaneously locates every pole of $\Gamma$ and proves it has no zeros. We route through the beta function, which repackages the product $\Gamma(z)\Gamma(1-z)$ as a single one-dimensional integral, evaluate that integral by a keyhole contour, and extend off the strip $0<\operatorname{Re}z<1$ by the identity theorem. The strategy is typical of the subject: establish an identity where two convergent objects overlap, then let continuation carry it everywhere both sides are meromorphic.

Step 1. Use the beta–gamma relation $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ with $x=z$ , $y=1-z$ . Since $x+y=1$ and $\Gamma(1)=1$ , this collapses to $B(z,1-z)=\Gamma(z)\Gamma(1-z)$ . Why: it converts the product of two gammas into one beta integral, trading a two-factor object for something we can integrate directly.

Step 2. Write the beta integral $B(z,1-z)=\int_0^1 t^{z-1}(1-t)^{-z}\,\mathrm{d}t$ and substitute $u=t/(1-t)$ , which maps $[0,1]$ onto $[0,\infty)$ and gives $B(z,1-z)=\int_0^\infty \dfrac{u^{z-1}}{1+u}\,\mathrm{d}u$ , valid for $0<\operatorname{Re}z<1$ . Why: the half-line form is the one a keyhole contour handles, because its only finite singularity is the simple pole at $u=-1$ .

Step 3. Evaluate by a keyhole contour around the branch cut on the positive real axis. The discontinuity of $u^{z-1}$ across the cut contributes a factor $(1-e^{2\pi i(z-1)})$ , and the residue at $u=-1=e^{i\pi}$ supplies $2\pi i\,e^{i\pi(z-1)}$ ; assembling them gives $\int_0^\infty u^{z-1}/(1+u)\,\mathrm{d}u=\pi/\sin\pi z$ . Why: the branch cut of the non-integer power is precisely what turns a residue into the trigonometric $\sin\pi z$ that carries the pole structure.

Step 4. Both sides of $\Gamma(z)\Gamma(1-z)=\pi/\sin\pi z$ are meromorphic on $\mathbb{C}$ and agree on the strip $0<\operatorname{Re}z<1$ , so by the identity theorem they agree everywhere. Reading it: the poles of $\Gamma(z)$ at $z=0,-1,\dots$ line up with poles of $\pi/\sin\pi z$ , and $\Gamma(1-z)$ is finite and nonzero there, so $\Gamma$ can never vanish. Why: analytic continuation upgrades a strip identity to a global one, and the never-vanishing right side is the clean proof that $1/\Gamma$ is entire. *As a check: at $z=\tfrac12$ , $\Gamma(\tfrac12)^2=\pi/\sin(\pi/2)=\pi$ , so $\Gamma(\tfrac12)=\sqrt{\pi}=1.77245$ ; and $\Gamma(\tfrac13)\Gamma(\tfrac23)=3.62760=\pi/\sin(\pi/3)$ .*

Go deeper · Derivation C · 1 minThe Legendre duplication formula

Derivation CThe Legendre duplication formula

The goal is to prove $\Gamma(z)\Gamma(z+\tfrac12)=2^{1-2z}\sqrt{\pi}\,\Gamma(2z)$ , the workhorse that manufactures every half-integer gamma value used in physics. Rather than manipulate integrals, we form the ratio of the two sides, show with the recurrence that it has period one, argue that periodicity plus controlled growth forces it to be constant, and pin the constant at one convenient point. This "show the ratio is constant" tactic is the cleanest route to any multiplication identity, and it generalizes verbatim to the full Gauss theorem.

Step 1. Define $\phi(z)=\dfrac{2^{2z-1}\,\Gamma(z)\,\Gamma(z+\tfrac12)}{\sqrt{\pi}\,\Gamma(2z)}$ ; the claim is $\phi\equiv1$ . Why: packaging both sides into one ratio converts the identity to be proved into the single statement " $\phi$ is the constant $1$ ".

Step 2. Apply $(2)$ to each gamma: $\Gamma(z+1)=z\Gamma(z)$ , $\Gamma(z+\tfrac32)=(z+\tfrac12)\Gamma(z+\tfrac12)$ , and $\Gamma(2z+2)=2z(2z+1)\Gamma(2z)$ . Substituting, the ratio $\phi(z+1)/\phi(z)=\dfrac{2^{2}\,z(z+\tfrac12)}{2z(2z+1)}=1$ , so $\phi$ has period one. Why: the powers of $2$ and the gamma recurrences cancel exactly, which is the arithmetic heart of the formula.

Step 3. $\phi$ is analytic and zero-free in the right half-plane (the gammas contribute no zeros, and their poles cancel between numerator and denominator), and Stirling's asymptotics bound its growth in a period strip. A periodic entire function bounded in a strip is bounded on all of $\mathbb{C}$ , so by Liouville it is constant, $\phi\equiv C$ . Why: periodicity turns a strip bound into a global bound, and Liouville's theorem then removes all $z$ -dependence.

Step 4. Fix $C$ at $z=\tfrac12$ : $\phi(\tfrac12)=\dfrac{2^{0}\,\Gamma(\tfrac12)\Gamma(1)}{\sqrt{\pi}\,\Gamma(1)}=\dfrac{\sqrt{\pi}}{\sqrt{\pi}}=1$ . Hence $\phi\equiv1$ , which rearranges to the boxed duplication formula. Why: one evaluation at a point where every gamma is elementary nails the constant and completes the proof. *As a check: at $z=\tfrac12$ both sides equal $\sqrt{\pi}$ ; the derived half-integer rule $\Gamma(n+\tfrac12)=(2n)!\,\sqrt{\pi}/(4^n n!)$ gives $\Gamma(\tfrac52)=\tfrac34\sqrt{\pi}=1.32934$ , matching direct evaluation.*

Limiting cases

The continued gamma function shows four qualitatively different behaviours across the plane; each is a check on the machinery above.

Case	Result	Meaning
Near the pole at $z=-n$	$\Gamma(z)\sim\dfrac{(-1)^n}{n!}\,\dfrac{1}{z+n}$	simple pole; the principal part dominates and the residue alternates in sign, shrinking factorially
Large $\lvert z\rvert$ (Stirling)	$\Gamma(z)\sim\sqrt{2\pi}\,z^{z-1/2}e^{-z}$ , $\lvert\arg z\rvert<\pi$	away from the negative real axis; gives factorial asymptotics and steepest-descent estimates
On vertical lines	$\lvert\Gamma(iy)\rvert^2=\dfrac{\pi}{y\sinh\pi y}$ , $\lvert\Gamma(\tfrac12+iy)\rvert^2=\dfrac{\pi}{\cosh\pi y}$	reflection forces exponential decay up the imaginary direction
Integer argument	$\Gamma(n+1)=n!$ , $\Gamma(1)=\Gamma(2)=1$	the continuation exactly reproduces the factorial it interpolates

Common trapThe poles come from the integral blowing up, and Γ must have zeros mirroring them

Both halves of this are wrong. The Euler integral does not "blow up" for $\operatorname{Re}z\le0$ — it simply fails to exist, because $\int_0^1 t^{x-1}\,\mathrm{d}t$ diverges at $t=0$ when $x\le0$ ; there is no function there until you continue. The poles are then created by the continuation, through the division by $z(z+1)\cdots(z+n)$ in $(3)$ , not inherited from the integral. And $\Gamma$ has no zeros anywhere: reflection forces $\Gamma(z)\Gamma(1-z)=\pi/\sin\pi z$ , whose finite right side can never make $\Gamma$ vanish. Students often expect zeros mirroring the poles; instead it is $1/\Gamma$ that is entire, with simple zeros at the non-positive integers. Keeping straight which function is entire — $1/\Gamma$ , not $\Gamma$ — is the single most useful correction here.

Worked examples

1 · Marching Γ into the left half-plane

Use the recurrence to evaluate $\Gamma$ at the negative half-integers $z=-\tfrac12,-\tfrac32,-\tfrac52$ , and check the pole at $z=-2$ .

Step 1. Anchor at $\Gamma(\tfrac12)=\sqrt{\pi}=1.77245$ from the reflection formula. Rearranging $(2)$ as $\Gamma(z)=\Gamma(z+1)/z$ steps one unit left each time.

Step 2. $\Gamma(-\tfrac12)=\Gamma(\tfrac12)/(-\tfrac12)=-2\sqrt{\pi}=\mathbf{-3.54491}$ . The sign flips because we divided by a negative number — the first hint of the alternating pole structure.

Step 3. $\Gamma(-\tfrac32)=\Gamma(-\tfrac12)/(-\tfrac32)=(-2\sqrt{\pi})/(-\tfrac32)=\tfrac43\sqrt{\pi}=\mathbf{2.36327}$ , and once more $\Gamma(-\tfrac52)=\Gamma(-\tfrac32)/(-\tfrac52)=-\tfrac{8}{15}\sqrt{\pi}=\mathbf{-0.94531}$ . The values alternate in sign and shrink, tracking the residues.

Step 4. Check the pole at $z=-2$ : near it $\Gamma(z)\approx(\tfrac12)/(z+2)$ using residue $(-1)^2/2!=\tfrac12$ . At $z=-1.9$ this predicts $0.5/0.1=\mathbf{5.0}$ ; direct evaluation gives $\Gamma(-1.9)=5.5635$ , the difference being the regular (non-singular) part of the Laurent series. The half-integers give $-2\sqrt{\pi},\ \tfrac43\sqrt{\pi},\ -\tfrac{8}{15}\sqrt{\pi}$ , and the residue $+\tfrac12$ correctly dominates near $z=-2$ .

2 · The volume of a unit n-ball

The volume of the unit ball in $n$ dimensions is $V_n=\pi^{n/2}/\Gamma(\tfrac n2+1)$ . Evaluate it for $n=1,2,3,5$ using half-integer gamma values, and confirm the familiar low-dimensional answers.

Step 1. The half-integer values come from the duplication formula: $\Gamma(\tfrac32)=\tfrac12\sqrt{\pi}=0.88623$ , $\Gamma(\tfrac52)=\tfrac34\sqrt{\pi}=1.32934$ , $\Gamma(\tfrac72)=\tfrac{15}{8}\sqrt{\pi}=3.32335$ , while $\Gamma(2)=1$ .

Step 2. For $n=1$ : $V_1=\pi^{1/2}/\Gamma(\tfrac32)=\sqrt{\pi}/(\tfrac12\sqrt{\pi})=\mathbf{2}$ — the length of the interval $[-1,1]$ . For $n=2$ : $V_2=\pi/\Gamma(2)=\pi/1=\boldsymbol{\pi}$ — the area of the unit disk.

Step 3. For $n=3$ : $V_3=\pi^{3/2}/\Gamma(\tfrac52)=5.56833/1.32934=4.18879=\mathbf{\tfrac43\pi}$ — the classic sphere volume. For $n=5$ : $V_5=\pi^{5/2}/\Gamma(\tfrac72)=17.4934/3.32335=5.26379=\mathbf{\tfrac{8\pi^2}{15}}$ .

Step 4. Only the continuation makes this possible: odd $n$ requires gamma at half-integers, which live between the factorial values and exist only because $\Gamma$ is defined off the integers. $V_1=2,\ V_2=\pi,\ V_3=\tfrac43\pi,\ V_5=\tfrac{8\pi^2}{15}$ — the even/odd split is the visible fingerprint of half-integer gamma.

3 · The 1/ε pole of dimensional regularization

A one-loop Feynman integral in $d=4-2\varepsilon$ dimensions produces a factor $\Gamma(2-\tfrac d2)=\Gamma(\varepsilon)$ . Expand it about $\varepsilon=0$ and identify the ultraviolet divergence.

Step 1. $\Gamma(\varepsilon)$ has its $n=0$ pole at the origin with residue $+1$ , so its Laurent series begins with $1/\varepsilon$ . Writing $\Gamma(\varepsilon)=\Gamma(1+\varepsilon)/\varepsilon$ and expanding the entire function $\Gamma(1+\varepsilon)=1-\gamma\varepsilon+(\tfrac{\gamma^2}{2}+\tfrac{\pi^2}{12})\varepsilon^2+\cdots$ gives the series.

Step 2. Dividing by $\varepsilon$ : $\Gamma(\varepsilon)=\dfrac1\varepsilon-\gamma+\Big(\dfrac{\gamma^2}{2}+\dfrac{\pi^2}{12}\Big)\varepsilon+\cdots$ . Numerically the linear coefficient is $\tfrac{\gamma^2}{2}+\tfrac{\pi^2}{12}=0.98906$ .

Step 3. The $\mathbf{1/\varepsilon}$ term is the ultraviolet divergence of the loop; the finite $-\gamma$ and the $\pi^2/12$ feed the renormalization counterterms — which is why $\gamma$ and $\pi^2$ litter every textbook loop calculation.

Step 4. Sanity check at $\varepsilon=0.1$ : the truncated series gives $10-0.57722+0.98906(0.1)=9.52169$ , against the exact $\Gamma(0.1)=9.51351$ — agreement to better than $0.1\%$ . The simple pole of $\Gamma$ at the origin is the $1/\varepsilon$ UV divergence, with residue exactly $1$ .

Physics of the portrait

The gamma function is where four centuries of analysis converge: a factorial that Euler taught to breathe between the integers, a symbol Legendre standardized, a product Weierstrass canonized, and a pole physicists now mine for infinities. The "portrait" here is the function itself.

Origins · 1 mingamma function and how this was discovered

gamma function

Γ(z), the meromorphic interpolation of n! with poles at 0, -1, -2, … · analytic continuation of special functions

The gamma function is one of the oldest objects in analysis and a template for everything that came after it. Leonhard Euler, answering a 1720s challenge to interpolate the factorial, produced both an infinite product and, by 1729–1730, the integral that becomes today's $\int_0^\infty t^{z-1}e^{-t}\,\mathrm{d}t$ — a single formula giving $n!$ a meaning at every real argument. For a century it was "the factorial function"; Adrien-Marie Legendre fixed the notation $\Gamma$ and the shift $\Gamma(n+1)=n!$ in his Exercices de calcul intégral around 1811, and stated the duplication formula that still generates half-integer values. Carl Friedrich Gauss, studying the hypergeometric series in 1812, proved the general multiplication theorem. The decisive analytic step was Karl Weierstrass's 1856 canonical product for $1/\Gamma$ , exhibiting it as an entire function over its zeros and making the constant $\gamma$ intrinsic to the theory. Harald Bohr and Johannes Mollerup closed the circle in 1922: log-convexity, with $\Gamma(1)=1$ and the recurrence, singles out $\Gamma$ uniquely among all interpolations of the factorial. The modern payoff arrived in physics — when Gerard 't Hooft and Martinus Veltman built dimensional regularization in 1972, the ultraviolet infinities of quantum field theory turned out to be nothing but the poles of $\Gamma$ in the spacetime dimension.

History

1729

— Euler interpolates the factorial with an integral, defining Γ for real arguments

1811

— Legendre fixes the notation Γ and states the duplication formula

1812

— Gauss proves the multiplication theorem and studies Γ on the line

1856

— Weierstrass writes 1/Γ as a canonical product over the negative integers

1972

— 't Hooft and Veltman exploit the pole of Γ(ε) in dimensional regularization

Going further

Texts. Arfken & Weber, Mathematical Methods for Physicists, Ch. 13 (Gamma Function) · Hassani, Mathematical Physics, Ch. on the gamma and beta functions · Stone & Goldbart, Mathematics for Physics, gamma-function chapter · Whittaker & Watson, A Course of Modern Analysis, Ch. 12 · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, Ch. 18 · Abramowitz & Stegun, Handbook of Mathematical Functions, Ch. 6. Key numerical anchors: integration by parts gives $\Gamma(z+1)=z\Gamma(z)$ ; residues from equation $(3)$ are $(-1)^n/n!$ , with $\Gamma(-\tfrac12)=-2\sqrt{\pi}$ and $\Gamma(-\tfrac32)=\tfrac43\sqrt{\pi}$ confirming the recurrence; the reflection formula rests on $\int_0^\infty u^{z-1}/(1+u)\,\mathrm{d}u=\pi/\sin\pi z$ , with checks $\Gamma(\tfrac13)\Gamma(\tfrac23)=3.62760$ and $\Gamma(\tfrac12)=\sqrt{\pi}$ ; duplication at $z=\tfrac12$ gives $\Gamma(\tfrac52)=\tfrac34\sqrt{\pi}=1.32934$ ; the $n$ -ball volumes are $V_3=\tfrac43\pi$ and $V_5=\tfrac{8\pi^2}{15}$ ; and the Laurent expansion is $\Gamma(\varepsilon)=1/\varepsilon-\gamma+0.98906\,\varepsilon+\cdots$ with $\gamma\approx0.5772156649$ .

Prerequisites. The Gamma Function · Recurrence Relations for Special Functions · The Beta Function. Leads to. Bessel Functions of the First and Second Kind · Spherical Bessel Functions · Legendre Polynomials · Hankel Functions and Asymptotic Behaviour of Bessel Functions.

Ask the tutor: ask me to walk you through continuing $\Gamma$ past any specific pole, or to show exactly how its $1/\varepsilon$ pole becomes a divergence in a real loop integral.

Why is the analytic continuation of Γ unique?

Because analytic functions are extraordinarily rigid. The identity theorem says two functions analytic on a connected domain that agree on any set with a limit point — even a tiny arc — must agree everywhere both are defined. The Euler integral is analytic on the whole right half-plane, so any meromorphic function on $\mathbb{C}$ matching it there is forced to match the continuation everywhere; there is no freedom to "choose" a different extension. This is why we speak of the gamma function, and it is the same rigidity that makes analytic continuation a legitimate tool across mathematical physics.

If the integral does not exist for Re z ≤ 0, where do the poles come from?

They are manufactured by the continuation, not present in the integral. For $\operatorname{Re}z\le0$ the Euler integral genuinely does not exist — there is no value to be singular. The recurrence $\Gamma(z)=\Gamma(z+n+1)/[z(z+1)\cdots(z+n)]$ supplies the value instead, and its denominator vanishes at $z=0,-1,\dots,-n$ . Dividing a finite, nonzero numerator by something going to zero is exactly what a simple pole is. So the poles are a feature of the extension mechanism: they sit at the integers because that is where the shifting denominator hits zero, and their residues $(-1)^n/n!$ come straight off the same formula.

Does Γ ever equal zero?

No — never, anywhere in the complex plane. The cleanest proof is the reflection formula $\Gamma(z)\Gamma(1-z)=\pi/\sin\pi z$ : the right-hand side is finite and nonzero except at integers, so neither factor on the left can vanish for non-integer $z$ ; at the integers one factor has a pole while the other is a finite nonzero factorial, so still no zero. Consequently $1/\Gamma(z)$ is an entire function — analytic on all of $\mathbb{C}$ with no poles — whose only zeros are the simple ones at $z=0,-1,-2,\dots$ inherited from the poles of $\Gamma$ . Remembering that $1/\Gamma$ , not $\Gamma$ , is the entire object prevents a very common error.

What is the Euler–Mascheroni constant doing inside the gamma function?

The constant $\gamma\approx0.5772$ appears wherever the growth of $\Gamma$ must be tamed. In the Weierstrass product $1/\Gamma(z)=z\,e^{\gamma z}\prod_n(1+z/n)e^{-z/n}$ , the bare product $\prod(1+z/n)$ diverges; the compensating factors $e^{-z/n}$ rescue convergence, and collecting them across all $n$ produces the overall $e^{\gamma z}$ , since $\gamma=\lim_{N\to\infty}(\sum_{n=1}^N 1/n-\ln N)$ . Equivalently $\gamma=-\Gamma'(1)$ , so it is the slope of $\ln\Gamma$ at $z=1$ . This is also why $\gamma$ turns up in the Laurent expansion $\Gamma(\varepsilon)=1/\varepsilon-\gamma+\cdots$ and hence throughout renormalized loop integrals in physics.

How does the duplication formula give half-integer factorials?

Set $z=\tfrac12$ in $\Gamma(z)\Gamma(z+\tfrac12)=2^{1-2z}\sqrt{\pi}\,\Gamma(2z)$ as a starting anchor, then use the recurrence to climb. More directly, the formula rearranges into the closed form $\Gamma(n+\tfrac12)=(2n)!\,\sqrt{\pi}/(4^n n!)$ , which for $n=0,1,2$ gives $\sqrt{\pi},\ \tfrac12\sqrt{\pi},\ \tfrac34\sqrt{\pi}$ . These are exactly the values physics needs: the volume of an odd-dimensional ball, the normalization of half-integer-order Bessel functions, and the small- $x$ behaviour of spherical Bessel functions all reduce to a $\Gamma(\ell+\tfrac32)$ . Without a formula relating $\Gamma(z)$ and $\Gamma(2z)$ , half-integer arguments would have no elementary closed form.

Why does the gamma function show up all over quantum field theory?

Because loop integrals in $d$ spacetime dimensions naturally evaluate to gamma functions of $d$ . Momentum integrals of the form $\int \mathrm{d}^d k\,(k^2+\Delta)^{-a}$ produce ratios like $\Gamma(a-\tfrac d2)/\Gamma(a)$ , and setting $d=4-2\varepsilon$ turns the ultraviolet divergence into a pole of $\Gamma$ at $\varepsilon\to0$ . The residue is the divergent coefficient a counterterm must cancel, and the finite Laurent terms — carrying $\gamma$ and $\pi^2$ — are the constants of renormalization. Dimensional regularization works precisely because $\Gamma$ is meromorphic in the dimension: the continuation built here is what lets "four dimensions minus a little bit" mean something.