Beta Function

The beta function is the second of Euler's two great definite integrals, the constant partner of the gamma function. Where the gamma function integrates one power against a decaying exponential over the half-line, the beta function integrates a product of two power laws over the unit interval, $B(x,y)=\int_0^1 t^{x-1}(1-t)^{y-1}\,\mathrm{d}t$ . The two exponents pull independently on the two endpoints — small $x$ concentrates weight at $t=0$ , small $y$ at $t=1$ — and the swap $t\mapsto 1-t$ makes the symmetry $B(x,y)=B(y,x)$ visible at a glance. This single integral is the cleanest tool a physicist owns for evaluating a whole zoo of definite integrals: powers, trigonometric powers, and rational substitutions all collapse onto it. Its power comes from one identity. The beta function is not an independent special function but a ratio of gammas, $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ , proved by turning a product of two gamma integrals into polar-like coordinates on the quarter plane. That identity does two jobs: it reduces any beta integral to gamma values you already know, and it hands the beta function the gamma function's analytic continuation for free — the convergent integral lives only where $\operatorname{Re}x>0$ and $\operatorname{Re}y>0$ , but the gamma ratio extends it meromorphically to the whole complex plane, with simple poles wherever $x$ , $y$ , or $-(x+y)$ meets a non-positive integer. For the working physicist the beta function is a normalization engine and a pole-counting device. It fixes the constant of the beta distribution, supplies the moment integrals of Jacobi-type orthogonal polynomials, converts one-loop Feynman parameter integrals into ratios of gammas in dimensional regularization, and — most spectacularly — is the Veneziano scattering amplitude, whose poles in the Mandelstam variables are an infinite tower of hadronic resonances. Reading those poles was the spark that lit string theory.

Key takeaways

The beta function is the Eulerian integral $B(x,y)=\int_0^1 t^{x-1}(1-t)^{y-1}\,\mathrm{d}t$ , convergent only for $\operatorname{Re}x>0$ and $\operatorname{Re}y>0$ .
Its master identity writes it as a ratio of gammas, $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ , so beta values reduce instantly to gamma values and inherit the gamma poles.
It is symmetric, $B(x,y)=B(y,x)$ , and wears three interchangeable disguises: a unit-interval integral, a trigonometric integral, and a half-line integral.
Beta integrals normalize the beta distribution, evaluate Wallis-type trigonometric integrals, build $n$ -sphere volumes, and tame loop integrals in dimensional regularization.
The Veneziano amplitude $A(s,t)=B(-\alpha(s),-\alpha(t))$ turns the beta function's gamma poles into an evenly spaced spectrum of resonances — the seed of string theory.
Anchor values worth memorizing: $B(1,1)=1$ , $B(1,y)=1/y$ , and $B(\tfrac12,\tfrac12)=\pi$ , the last equivalent to $\Gamma(\tfrac12)=\sqrt{\pi}$ .
Key idea: any integral of the form (power)×(power) over a finite interval is a disguised beta function, and $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ converts it instantly into gamma functions whose values and poles you already know.

Where this lives today

The beta function is quietly load-bearing across modern physics and data science. In precision collider physics every one-loop master integral is evaluated in dimensional regularization by reducing Feynman-parameter integrals to the half-line beta form; the resulting gamma ratios expose ultraviolet divergences as poles in $\varepsilon=(4-d)/2$ , and expanding those gammas turns a raw loop diagram into a finite prediction tested against LHC data. In statistics the beta distribution — normalized by exactly this function — is the conjugate prior for a binomial probability, so every Bayesian A/B test and beta-binomial model computes beta functions under the hood. In statistical mechanics the $n$ -sphere volumes built from beta integrals set the phase-space measure for $d$ -dimensional momentum integrals and the density of states. And the Veneziano amplitude remains the anchor of string theory: its descendants — the Virasoro–Shapiro amplitude and the whole industry of string scattering amplitudes — are generalized Euler integrals whose poles are still being read as particles.

The physical picture

Picture the unit interval as a thin rod and the integrand $t^{x-1}(1-t)^{y-1}$ as a mass density laid along it. The factor $t^{x-1}$ controls the left end: if $x<1$ the density blows up at $t=0$ , if $x>1$ it vanishes there. The factor $(1-t)^{y-1}$ does the mirror job at the right end. The beta function $B(x,y)$ is just the total mass of the rod — the area under the density. Divide that density by its own total mass and you get a genuine probability distribution, the beta distribution, whose entire shape is dialed by the two exponents. The first figure shows three of them and makes the endpoint control concrete.

Figure

Three beta densities t^(x−1)(1−t)^(y−1)/B(x,y) on the unit interval. The (2,2) curve is symmetric with mean 1/2; (5,2) piles its weight toward the right endpoint with mode 0.8 and mean 5/7; (2,5) is its mirror with mean 2/7. The two exponents pull independently on the two ends, and B(x,y) — here B(2,2)=1/6, B(5,2)=1/30 — is exactly the area that normalizes each curve to total probability 1.

The second reading is the one physicists reach for. Because $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ is a ratio of gammas, and the gamma function has a simple pole at every non-positive integer, the beta function inherits an infinite grid of poles. In the Veneziano amplitude the beta arguments become linear functions of the collision energy through a Regge trajectory $\alpha(s)=\alpha_0+\alpha' s$ , so each gamma pole becomes a particle — a resonance at a definite mass. The evenly spaced ladder below is the whole hadronic tower, and its existence inside one beta function is why a line of mathematics grew into string theory.

Figure

The Veneziano amplitude A(s,t) = B(−α(s), −α(t)) inherits the gamma poles: it blows up wherever α(s) = 0, 1, 2, 3, …. On a linear Regge trajectory α(s) = α₀ + α′s each pole is a resonance of mass M_n² = s_n = (n − α₀)/α′, an evenly spaced tower with spacing Δ(M²) = 1/α′. Here α₀ = −0.5 and α′ = 1, so the resonances sit at s = 0.5, 1.5, 2.5, 3.5.

The core equations

The definition is a single convergent integral. Two power-law factors on the unit interval, one exponent per endpoint; the whole content of the beta function is contained here, and every later form is a change of variable away.

B(x,y)=\int_0^1 t^{x-1}(1-t)^{y-1}\,\mathrm{d}t,\qquad \operatorname{Re}x>0,\ \operatorname{Re}y>0 \qquad (1)

where: $x,y$ — complex parameters, each with strictly positive real part; $t$ — integration variable running across the unit interval. Reading: the constraint is not decorative — at $t=0$ the integrand behaves like $t^{x-1}$ , integrable only if $\operatorname{Re}x>0$ , and at $t=1$ like $(1-t)^{y-1}$ , integrable only if $\operatorname{Re}y>0$ . The master identity dissolves the beta function into gammas. It is the reason the beta function is a convenience rather than a new transcendental, and the bridge that carries the gamma function's poles and analytic continuation over to $B$ .

B(x,y)=\frac{\Gamma(x)\,\Gamma(y)}{\Gamma(x+y)} \qquad (2)

where: $\Gamma$ — the gamma function, $\Gamma(z)=\int_0^\infty t^{z-1}e^{-t}\,\mathrm{d}t$ for $\operatorname{Re}z>0$ , continued meromorphically elsewhere. Reading: any beta value reduces to three gamma evaluations, and $B$ has a simple pole wherever $x$ , $y$ , or $-(x+y)$ is a non-positive integer — but never a zero, since $\Gamma$ has none. The symmetry is the fingerprint of the substitution $t\mapsto 1-t$ , which swaps the two endpoints and therefore the two exponents.

B(x,y)=B(y,x) \qquad (3)

where: $B$ — the beta function. Reading: geometrically the density is reflected about the midpoint $t=\tfrac12$ , which cannot change the total area; algebraically it is the manifest symmetry of $\Gamma(x)\Gamma(y)/\Gamma(x+y)$ under $x\leftrightarrow y$ . The trigonometric form is what the beta function looks like on a quarter-circle instead of a segment. It is the single most useful disguise in physics, because angular integrals over spheres and the Wallis powers of sine and cosine are exactly of this shape.

B(x,y)=2\int_0^{\pi/2}\sin^{2x-1}\theta\,\cos^{2y-1}\theta\,\mathrm{d}\theta \qquad (4)

where: $\theta$ — polar angle, with $t=\sin^2\theta$ so that $1-t=\cos^2\theta$ . Reading: matching a target integral $\int_0^{\pi/2}\sin^m\theta\cos^n\theta\,\mathrm{d}\theta$ against this fixes $x=\tfrac{m+1}{2}$ , $y=\tfrac{n+1}{2}$ , and the answer drops out as a ratio of gammas. The half-line form maps the unit interval onto $[0,\infty)$ . It is the shape that appears when a Feynman parameter or a loop momentum has already been rendered as a single rational integral over the positive axis.

B(x,y)=\int_0^\infty \frac{u^{x-1}}{(1+u)^{x+y}}\,\mathrm{d}u \qquad (5)

where: $u$ — half-line variable, with $t=u/(1+u)$ . Reading: in dimensional regularization the exponents become linear in the spacetime dimension $d$ , so the gamma ratio on the right carries the ultraviolet divergence as a pole in $\varepsilon=(4-d)/2$ . The recurrence is the beta function's echo of $\Gamma(z+1)=z\,\Gamma(z)$ . It walks integer-shifted arguments up or down and turns families of beta integrals into simple products.

B(x+1,y)=\frac{x}{x+y}\,B(x,y),\qquad B(1,y)=\frac1y \qquad (6)

where: $x,y$ — parameters with positive real part; the anchor $B(1,y)=1/y$ comes from $\int_0^1(1-t)^{y-1}\,\mathrm{d}t$ . Reading: repeated use collapses any $B$ with integer arguments to a rational number, and it is the identity behind the moments of the beta distribution.

Derivation AThe beta gamma identity from a Gaussian double integral

The goal is to prove the single most important fact about the beta function, that $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ , starting from nothing but the integral definitions of the two Eulerian integrals. The trick is to write the product $\Gamma(x)\Gamma(y)$ as one double integral over the positive quadrant and change to a radial variable $s=u+v$ and a fractional variable $t=u/(u+v)$ that separates the two factors. The double integral then factorizes into a gamma piece and a beta piece, and rearranging gives the identity every other result here leans on.

Step 1. Write each gamma as its integral and multiply, $\Gamma(x)\Gamma(y)=\int_0^\infty\!\int_0^\infty u^{x-1}v^{y-1}e^{-(u+v)}\,\mathrm{d}u\,\mathrm{d}v$ . Why: a product of two one-dimensional integrals is literally a two-dimensional integral over the quarter plane, and the shared exponential $e^{-(u+v)}$ is begging for the sum $u+v$ to become a new variable.

Step 2. Change variables to $u=s\,t$ , $v=s(1-t)$ with $s=u+v\in[0,\infty)$ and $t=u/(u+v)\in[0,1]$ ; the Jacobian is $\partial(u,v)/\partial(s,t)=s$ , so $\mathrm{d}u\,\mathrm{d}v=s\,\mathrm{d}s\,\mathrm{d}t$ . Why: $s$ measures the total and $t$ the split between the two axes, exactly the pair that turns the coupled integrand into a product of a pure- $s$ factor and a pure- $t$ factor.

Step 3. Substitute: $u^{x-1}v^{y-1}=s^{x+y-2}t^{x-1}(1-t)^{y-1}$ and $e^{-(u+v)}=e^{-s}$ , so the integral becomes $\int_0^\infty s^{x+y-1}e^{-s}\,\mathrm{d}s\int_0^1 t^{x-1}(1-t)^{y-1}\,\mathrm{d}t$ . Why: the powers of $s$ collect to $s^{x+y-2}\cdot s=s^{x+y-1}$ , and because the region is the full rectangle $s\in[0,\infty)$ , $t\in[0,1]$ , the double integral genuinely factorizes into two independent one-dimensional integrals.

Step 4. Recognize the two factors — the $s$ -integral is $\Gamma(x+y)$ , the $t$ -integral is $B(x,y)$ — so $\Gamma(x)\Gamma(y)=\Gamma(x+y)\,B(x,y)$ , and dividing gives $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ . Why: the right-hand side is now defined for all complex $x,y$ by the meromorphic gamma function, which is precisely how the beta function is continued beyond the strip where its own integral converges. *Verified: at $x=y=\tfrac12$ the identity gives $\Gamma(\tfrac12)^2/\Gamma(1)=\pi$ , matching the direct integral $\int_0^1[t(1-t)]^{-1/2}\mathrm{d}t=\pi$ ; at $x=4,y=3$ it gives $\Gamma(4)\Gamma(3)/\Gamma(7)=6\cdot2/720=1/60$ , matching the elementary $\int_0^1 t^3(1-t)^2\,\mathrm{d}t$ .*

Go deeper · Derivation B · 1 minThe trigonometric form and the diagonal value pi

Derivation BThe trigonometric form and the diagonal value pi

The aim is to derive the half-angle representation of the beta function and use it to pin down the diagonal value $B(\tfrac12,\tfrac12)=\pi$ , which is equivalent to the famous $\Gamma(\tfrac12)=\sqrt{\pi}$ and hence to the Gaussian integral. The move is a single substitution $t=\sin^2\theta$ carrying the unit interval onto the first quadrant of angle; there the beta function becomes an integral of powers of sine and cosine, and setting both exponents to zero leaves a bare $\int_0^{\pi/2}\mathrm{d}\theta$ . Reading the same value back through the gamma representation then delivers $\Gamma(\tfrac12)$ without ever touching a Gaussian.

Step 1. Substitute $t=\sin^2\theta$ , so $1-t=\cos^2\theta$ and $\mathrm{d}t=2\sin\theta\cos\theta\,\mathrm{d}\theta$ , with $t:0\to1$ mapping to $\theta:0\to\tfrac{\pi}{2}$ . Why: squaring the sine is what produces the clean $1-t=\cos^2\theta$ , so both power-law factors turn into powers of trigonometric functions simultaneously.

Step 2. Insert and collect powers: $B(x,y)=\int_0^{\pi/2}(\sin^2\theta)^{x-1}(\cos^2\theta)^{y-1}\,2\sin\theta\cos\theta\,\mathrm{d}\theta=2\int_0^{\pi/2}\sin^{2x-1}\theta\,\cos^{2y-1}\theta\,\mathrm{d}\theta$ . Why: the two extra factors from $\mathrm{d}t$ shift each exponent by one, converting $2x-2$ into $2x-1$ and $2y-2$ into $2y-1$ — the reason the trigonometric powers are odd-looking.

Step 3. Set $x=y=\tfrac12$ : both exponents $2x-1$ and $2y-1$ vanish, so $B(\tfrac12,\tfrac12)=2\int_0^{\pi/2}\mathrm{d}\theta=2\cdot\tfrac{\pi}{2}=\pi$ . Why: the delicate endpoint singularities of the original integrand $[t(1-t)]^{-1/2}$ are completely regularized by the substitution — the answer is just the length of the angular interval.

Step 4. Compare with the gamma form: $B(\tfrac12,\tfrac12)=\Gamma(\tfrac12)^2/\Gamma(1)=\Gamma(\tfrac12)^2$ , so $\Gamma(\tfrac12)^2=\pi$ and $\Gamma(\tfrac12)=\sqrt{\pi}$ . Why: this is the Gaussian integral $\int_{-\infty}^\infty e^{-t^2}\mathrm{d}t=\sqrt\pi$ arriving through the back door, showing the beta and gamma functions share the same $\sqrt\pi$ at their common half-integer. *Verified: numerically $B(\tfrac12,\tfrac12)=3.14159\ldots=\pi$ and $\Gamma(\tfrac12)=1.77245\ldots=\sqrt\pi$ ; cross-checked against Abramowitz and Stegun 6.1.8.*

Go deeper · Derivation C · 1 minThe Veneziano amplitude as a beta function

Derivation CThe Veneziano amplitude as a beta function

The goal is to show that the four-point dual-resonance amplitude that launched string theory is literally an Euler beta function of the Mandelstam variables, $A(s,t)=B(-\alpha(s),-\alpha(t))$ , and that its poles — inherited from the gamma function — form an infinite, evenly spaced tower of resonances once the Regge trajectory is linear. We start from the integral form, read off where it diverges, translate that into a mass spectrum, and inspect a residue to confirm each pole carries the right tower of spins. This is the calculation in which counting the poles of a special function became counting the particles of a theory.

Step 1. Take the amplitude in its integral form $A(s,t)=\int_0^1 x^{-\alpha(s)-1}(1-x)^{-\alpha(t)-1}\,\mathrm{d}x=B(-\alpha(s),-\alpha(t))=\dfrac{\Gamma(-\alpha(s))\Gamma(-\alpha(t))}{\Gamma(-\alpha(s)-\alpha(t))}$ . Why: the swap $x\leftrightarrow 1-x$ exchanges the two exponents, which is the crossing (duality) symmetry between the $s$ -channel and the $t$ -channel — a single term already contains both.

Step 2. Locate the poles: $\Gamma(-\alpha(s))$ has simple poles wherever its argument is a non-positive integer, i.e. $-\alpha(s)=0,-1,-2,\ldots$ , which is $\alpha(s)=0,1,2,\ldots$ . Why: the gamma function's poles are the amplitude's poles, and in scattering a pole in the energy variable is the signature of an intermediate particle — a resonance.

Step 3. Feed in a linear trajectory $\alpha(s)=\alpha_0+\alpha' s$ ; the pole condition $\alpha(s)=n$ becomes $s_n=(n-\alpha_0)/\alpha'$ , an arithmetic progression of squared masses $M_n^2=s_n$ with uniform spacing $\Delta(M^2)=1/\alpha'$ . Why: a straight Regge trajectory is exactly what turns the integer ladder of gamma poles into an evenly spaced mass spectrum — the hallmark of a vibrating string, whose modes have energies stepping by a fixed quantum.

Step 4. Inspect the residue at $\alpha(s)=n$ : it is $\operatorname*{Res}A=\dfrac{(\alpha(t)+1)(\alpha(t)+2)\cdots(\alpha(t)+n)}{n!}$ , a degree- $n$ polynomial in $\alpha(t)$ . Why: a residue that is a polynomial of degree $n$ in the crossed variable means the resonance at level $n$ exchanges spins up to $n$ — the tower is not a single particle but a full multiplet, which is dual resonance made explicit. *Verified: pole positions and the polynomial residue structure reproduce the standard Veneziano-model treatment; the beta-to-gamma step is the identity proved in the first derivation.*

Limiting cases

Boundary and special-argument behaviour of $B(x,y)$ , each read off the gamma representation:

Case	Result	Meaning
Diagonal half-integers	$B\!\left(\tfrac12,\tfrac12\right)=\pi$	Since $\Gamma(\tfrac12)=\sqrt{\pi}$ and $\Gamma(1)=1$ ; equivalently $\int_0^1\frac{\mathrm{d}t}{\sqrt{t(1-t)}}=\pi$ .
Unit arguments	$B(1,1)=1,\qquad B(1,y)=\dfrac1y$	From $\int_0^1(1-t)^{y-1}\,\mathrm{d}t=1/y$ ; the anchor for the recurrence.
Large $y$ , fixed $x$	$B(x,y)\sim\Gamma(x)\,y^{-x}\quad(y\to\infty)$	From $\Gamma(y)/\Gamma(x+y)\sim y^{-x}$ (Stirling); the mass concentrates near $t=0$ .
Approaching a pole	$B(x,y)\to\infty$ as $x\to0,-1,-2,\dots$	A simple pole appears whenever $x$ , $y$ , or $x+y$ hits a non-positive integer, inherited from $\Gamma$ .
Symmetric integer pair	$B(m,n)=\dfrac{(m-1)!\,(n-1)!}{(m+n-1)!}$	Positive integers reduce the beta function to a ratio of factorials, e.g. $B(4,3)=\tfrac{3!\,2!}{6!}=\tfrac1{60}$ .

Common trapThe integral definition is not the function

It is tempting to treat $\int_0^1 t^{x-1}(1-t)^{y-1}\,\mathrm{d}t$ as the beta function everywhere, but that integral exists only in the open quadrant $\operatorname{Re}x>0,\ \operatorname{Re}y>0$ . Outside it the integral diverges — at $x=0$ the integrand already behaves like $t^{-1}$ near the origin, whose integral is logarithmically infinite. The beta function proper is the analytic continuation supplied by $\Gamma(x)\Gamma(y)/\Gamma(x+y)$ , and there it is not merely large but genuinely infinite: it has a simple pole. So $B(0,1)$ , $B(-1,3)$ , and $B(-2,-2)$ are not numbers to be computed from the integral but poles to be tracked. Conversely, the gamma ratio never vanishes — the beta function has poles but no zeros — a fact worth remembering before you try to solve $B(x,y)=0$ .

Worked examples

1 · A polynomial integral, three ways (fully worked)

Evaluate $\displaystyle I=\int_0^1 t^{3}(1-t)^{2}\,\mathrm{d}t$ using the beta function, cross-check by the recurrence, and cross-check again by brute expansion.

Step 1. Recognize the shape. Matching $t^{x-1}(1-t)^{y-1}$ with $t^3(1-t)^2$ gives $x-1=3$ and $y-1=2$ , so $x=4$ , $y=3$ and $I=B(4,3)$ . The integral is a beta function with $x=4,\ y=3$ .

Step 2. Use the identity. With positive integers the gamma ratio is a ratio of factorials, $B(4,3)=\dfrac{\Gamma(4)\Gamma(3)}{\Gamma(7)}=\dfrac{3!\,2!}{6!}=\dfrac{6\cdot2}{720}=\dfrac{12}{720}=\mathbf{\dfrac{1}{60}}$ .

Step 3. Cross-check by the recurrence $B(x+1,y)=\dfrac{x}{x+y}B(x,y)$ , starting from $B(1,3)=\tfrac13$ . Then $B(2,3)=\tfrac{1}{4}\cdot\tfrac13=\tfrac{1}{12}$ , $B(3,3)=\tfrac{2}{5}\cdot\tfrac1{12}=\tfrac{1}{30}$ , and $B(4,3)=\tfrac{3}{6}\cdot\tfrac1{30}=\tfrac{1}{60}$ . The recurrence reproduces $1/60$ .

Step 4. Cross-check by brute force. Expand $(1-t)^2=1-2t+t^2$ , so $I=\int_0^1(t^3-2t^4+t^5)\,\mathrm{d}t=\tfrac14-\tfrac25+\tfrac16$ . Over the common denominator $60$ : $\tfrac{15}{60}-\tfrac{24}{60}+\tfrac{10}{60}=\tfrac{1}{60}$ . All three routes agree: $I=1/60$ . The beta function turned an expansion into a one-line factorial ratio — a saving that grows explosively as the exponents rise.

2 · A Wallis integral and the volume of the unit 3-ball

Use the trigonometric form to evaluate $\displaystyle J=\int_0^{\pi/2}\sin^{4}\theta\,\cos^{2}\theta\,\mathrm{d}\theta$ , then show how the same machinery builds the volume $V_3=\tfrac43\pi$ of the unit 3-ball.

Step 1. Match exponents. Comparing $\sin^m\theta\cos^n\theta$ with $m=4,\ n=2$ against equation (4) gives $x=\tfrac{m+1}{2}=\tfrac52$ and $y=\tfrac{n+1}{2}=\tfrac32$ , and the leading factor of two means $J=\tfrac12 B(\tfrac52,\tfrac32)$ .

Step 2. Evaluate the beta value. $B(\tfrac52,\tfrac32)=\dfrac{\Gamma(\tfrac52)\Gamma(\tfrac32)}{\Gamma(4)}$ . With $\Gamma(\tfrac52)=\tfrac34\sqrt\pi$ , $\Gamma(\tfrac32)=\tfrac12\sqrt\pi$ and $\Gamma(4)=6$ , this is $\dfrac{(\tfrac34\sqrt\pi)(\tfrac12\sqrt\pi)}{6}=\dfrac{3\pi/8}{6}=\dfrac{\pi}{16}$ , so $J=\tfrac12\cdot\dfrac{\pi}{16}=\mathbf{\dfrac{\pi}{32}}$ . This matches the double-factorial rule $J=\dfrac{3!!\,1!!}{6!!}\cdot\dfrac{\pi}{2}=\dfrac{3}{48}\cdot\dfrac{\pi}{2}=\dfrac{\pi}{32}$ .

Step 3. Scale up to geometry. The volume of the unit $n$ -ball is $V_n=\pi^{n/2}/\Gamma(\tfrac n2+1)$ , a formula assembled by exactly these angular beta integrals. For $n=3$ , $V_3=\dfrac{\pi^{3/2}}{\Gamma(\tfrac52)}=\dfrac{\pi^{3/2}}{\tfrac34\sqrt\pi}=\dfrac{4\pi}{3}$ . The beta function reproduces the familiar $V_3=\tfrac43\pi$ , and the same construction gives $V_2=\pi$ and $V_4=\tfrac12\pi^2$ — the beta function builds the geometry of every dimension.

3 · Reading a resonance off the Veneziano amplitude

With trajectory $\alpha(s)=\tfrac12+s$ (so $\alpha_0=\tfrac12$ , $\alpha'=1$ in units where $M^2$ and $s$ coincide), locate the first three resonances of $A(s,t)=B(-\alpha(s),-\alpha(t))$ in the $s$ -channel.

Step 1. The amplitude has poles at $\alpha(s)=0,1,2,\ldots$ . Solving $\tfrac12+s=n$ gives $s_n=n-\tfrac12$ , so the first three sit at $s_0=-\tfrac12$ , $s_1=\tfrac12$ , $s_2=\tfrac32$ . The masses step by $\Delta(M^2)=1/\alpha'=1$ , evenly spaced as the linear trajectory demands.

Step 2. Interpret. The lowest pole at $s_0=-\tfrac12$ is a tachyon (negative $M^2$ ), an artifact of the bosonic model that later string constructions remove; the physical content is the even spacing of the tower above it — the signature that a single beta function describes an infinite family of particles, not one. This is why the amplitude is remembered as the accidental discovery of a string spectrum inside a special function.

Physics of the portrait

The beta function is a case study in how eighteenth-century integral calculus becomes, two centuries later, the seed of a physical theory. Euler wrote it down as a curiosity of interpolation; Legendre organized it; and in 1968 a young theorist recognized its poles as particles.

Origins · 1 minbeta function and how this was discovered

beta function

Euler's second Eulerian integral, from 1730 to the Veneziano amplitude · special functions and dual-resonance models

The beta function begins with Leonhard Euler, who in the years around 1730 was hunting for a way to interpolate the factorial to non-integer arguments — the problem that produced the gamma function — and in the same circle of ideas met its companion, the integral of a product of two powers over the unit interval. Euler's Institutionum calculi integralis (1768) develops these "Eulerian integrals" and the reflection and reduction relations among them. Adrien-Marie Legendre, in his Exercices de calcul intégral (1811), gave the subject its lasting vocabulary and notation: he fixed the symbol $\Gamma$ and christened the two integrals the Eulerian integrals of the first kind (the beta) and the second kind (the gamma). Jacques Binet's 1839 memoir systematized their theory further, and the modern symbol $B(p,q)$ settled into use through the nineteenth-century French tradition. For a hundred and thirty years the beta function was a workhorse of definite-integral tables and nothing more. Then in 1968 Gabriele Veneziano, seeking an amplitude for strong-interaction scattering that respected the observed crossing symmetry and Regge behaviour, wrote down $A(s,t)=B(-\alpha(s),-\alpha(t))$ — and its infinite ladder of gamma poles turned out to be a spectrum of hadronic resonances. Within two years Nambu, Nielsen, and Susskind had recognized that spectrum as the vibrational modes of a relativistic string. Euler's integral of two powers had become the first equation of string theory.

History

1730

— Euler introduces the interpolation integrals underlying the gamma and beta functions

1768

— Euler's Institutionum calculi integralis develops the beta integral and its reduction relations

1811

— Legendre names the Eulerian integrals of the first and second kind and fixes the Γ notation

1839

— Binet's memoir systematizes the theory of the Eulerian integrals

1968

— Veneziano writes the four-point dual-resonance amplitude as a beta function

Going further

Texts. Arfken and Weber, Mathematical Methods for Physicists, 7th ed., the Gamma–Beta chapter · Stone and Goldbart, Mathematics for Physics, the Special Functions chapter (Eulerian integrals) · Hassani, Mathematical Physics, gamma and beta sections · Riley, Hobson and Bence, Mathematical Methods for Physics and Engineering, Special Functions chapter · Abramowitz and Stegun, Handbook of Mathematical Functions, §6.2 · NIST Digital Library of Mathematical Functions, §5.12.

Provenance. The definition, the gamma identity, the trigonometric and half-line forms, the recurrence, and the value $B(\tfrac12,\tfrac12)=\pi$ follow Arfken–Weber and Stone–Goldbart. A verifier should recompute: (1) the Gaussian double-integral derivation of $B=\Gamma\Gamma/\Gamma$ ; (2) $B(4,3)=1/60$ by factorials, by recurrence, and by expansion; (3) $\int_0^{\pi/2}\sin^4\theta\cos^2\theta\,\mathrm{d}\theta=\pi/32$ and $V_3=4\pi/3$ ; (4) $\Gamma(\tfrac12)=\sqrt\pi$ from $B(\tfrac12,\tfrac12)=\pi$ (Abramowitz and Stegun 6.1.8). The Veneziano pole positions $s_n=(n-\alpha_0)/\alpha'$ and the polynomial residue structure are standard dual-resonance material.

Prerequisites. The Gamma Function · Analytic Continuation and Poles of the Gamma Function. Leads to. Orthogonal Polynomials: General Theory · Legendre Polynomials · Bessel Functions of the First and Second Kind.

Ask the tutor: Bring me any definite integral of powers — over the unit interval, a quarter circle, or the half-line — and I will show you the beta function hiding inside it and turn it into a ratio of gammas in one line.

Is the beta function really just the gamma function in disguise?

For the purpose of evaluation, yes. The identity $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ means the beta function contains no information the gamma function does not already carry, and its poles, its analytic continuation, and its special values are all inherited from $\Gamma$ . What the beta function adds is packaging: it is the natural form for integrals over a finite interval and for symmetric two-argument problems, where writing three separate gammas would obscure the structure. The Veneziano amplitude is the clearest example — presenting it as $B(-\alpha(s),-\alpha(t))$ makes the crossing symmetry between the $s$ - and $t$ -channels manifest, whereas the gamma ratio hides it. So the beta function is a viewpoint, not a new transcendental.

Why does the integral only converge for positive real parts?

Because of what happens at the two endpoints. Near $t=0$ the integrand behaves like $t^{x-1}$ , and $\int_0 t^{x-1}\,\mathrm{d}t$ converges only when $\operatorname{Re}x>0$ — at $x=0$ it is $\int t^{-1}$ , a logarithm that diverges. By the symmetric roles of the endpoints, the behaviour near $t=1$ demands $\operatorname{Re}y>0$ . Outside that quadrant the integral simply does not exist as a finite number, which is why the defining object beyond the strip is not the integral but its gamma-ratio continuation. The continued function is then genuinely infinite — a simple pole — at every non-positive integer value of $x$ , $y$ , or $-(x+y)$ , not merely undefined.

How is the beta function connected to probability?

Through normalization. The function $f(t)=t^{x-1}(1-t)^{y-1}/B(x,y)$ on the unit interval is a probability density — the beta distribution — precisely because dividing by $B(x,y)$ , the total area under the un-normalized integrand, forces $\int_0^1 f\,\mathrm{d}t=1$ . Its moments are ratios of beta functions, $\langle t^k\rangle=B(x+k,y)/B(x,y)$ , which the recurrence collapses to simple products; the mean is $x/(x+y)$ and the mode is $(x-1)/(x+y-2)$ . Because it is the conjugate prior of the binomial parameter, the beta distribution is the backbone of Bayesian inference for proportions — every time you update a belief about a success probability, a beta function is doing the bookkeeping.

What makes the Veneziano amplitude so historically important?

It was the accidental discovery that a special function's poles could be a particle spectrum. Veneziano was not trying to invent string theory; he wanted a strong-interaction amplitude with the right crossing symmetry and Regge asymptotics, and the beta function fit the bill. But its structure carried an unintended payload: an infinite tower of poles at evenly spaced masses, exactly what you would expect from the vibrational modes of a string. Within two years Nambu, Nielsen, and Susskind identified that string, and an entire framework grew from it. The beta function thus sits at a genuine hinge of physics — the point where a nineteenth-century integral became the first equation of a twentieth-century theory of everything.

When should I reach for the trigonometric form versus the half-line form?

Match the form to the geometry of your integral. If the integration variable is an angle and you are staring at powers of sine and cosine — angular integrals over spheres, radiation patterns, Wallis reductions — use the trigonometric form $B=2\int_0^{\pi/2}\sin^{2x-1}\theta\cos^{2y-1}\theta\,\mathrm{d}\theta$ and read off $x=\tfrac{m+1}{2}$ , $y=\tfrac{n+1}{2}$ . If instead your variable runs over the whole positive axis and appears as $u^{x-1}/(1+u)^{x+y}$ — Feynman parameters, propagator denominators, Mellin-type integrals — use the half-line form. Both are the same beta function; choosing the right disguise saves you the change of variables and lets you go straight to the gamma ratio.

Does the beta function have any zeros?

No. Since $B(x,y)=\Gamma(x)\Gamma(y)/\Gamma(x+y)$ and the gamma function is nowhere zero, a product of gammas in the numerator can never vanish, and dividing by $\Gamma(x+y)$ cannot create a zero either. What $B$ has instead is simple poles wherever a numerator gamma blows up — at $x=0,-1,-2,\ldots$ or $y=0,-1,-2,\ldots$ . The one subtlety is cancellation: when $x+y$ is also a non-positive integer the denominator gamma has a pole too, and the ratio can stay finite there, so a naive pole count must check whether a numerator pole is cancelled by the denominator. An actual zero of $B$ never occurs.