Gaussian Hypergeometric Function

The Gaussian hypergeometric function ₂F₁(a,b;c;z) is the single most important special function in mathematical physics — not because it appears more often than Bessel or Legendre functions, but because it contains them. It is defined by the most modest generalization of the geometric series: keep the power series $\sum c_n z^n$ , but let the ratio $c_{n+1}/c_n$ be an arbitrary rational function of the index $n$ rather than a constant. The geometric series has term ratio $1$ ; the hypergeometric series with two roots upstairs and one downstairs sits one rung higher, and that single extra degree of freedom is enough to sweep up almost every function a physicist meets. Underneath the series lives a differential equation, $z(1-z)w''+[c-(a+b+1)z]w'-ab\,w=0$ , and this is where the deep structure hides. It has regular singular points at exactly three places — $z=0$ , $z=1$ , and $z=\infty$ — and is regular everywhere else. Riemann proved the converse: any second-order linear ODE with precisely three regular singular points is, after a Möbius change of variable and a power-law rescaling, the hypergeometric equation. So studying ₂F₁ is studying all three-singular-point problems at once. Two levers then turn it into everything else — truncation ( $a=-n$ ) terminates the series into the classical orthogonal polynomials, and confluence (colliding $z=1$ with $z=\infty$ ) yields ₁F₁ and hence the Bessel, Laguerre and Coulomb functions. This article develops the series, the equation, the integral that continues it past its disk of convergence, and the transformations that make it computable everywhere.

Key takeaways

The Gaussian hypergeometric function $_2F_1(a,b;c;z)=\sum_{n\ge0}\frac{(a)_n(b)_n}{(c)_n}\frac{z^n}{n!}$ is the power series whose consecutive-term ratio is a rational function of the index $n$ .
It is the canonical solution near $z=0$ of the hypergeometric ODE $z(1-z)w''+[c-(a+b+1)z]w'-ab\,w=0$ — the most general second-order linear ODE with exactly three regular singular points, at $0$ , $1$ , $\infty$ .
The series converges only for $|z|<1$ ; Euler's integral and the connection formulae continue the function to the plane cut along $[1,\infty)$ .
At $z=1$ , Gauss's theorem gives $_2F_1(a,b;c;1)=\Gamma(c)\Gamma(c-a-b)/[\Gamma(c-a)\Gamma(c-b)]$ whenever $\mathrm{Re}(c-a-b)>0$ .
Truncation ( $a=-n$ ) yields the classical orthogonal polynomials; confluence (merging $1$ with $\infty$ ) yields Bessel, Laguerre and Coulomb functions.
Key idea: a power series whose term ratio is a rational function of the index $n$ is a hypergeometric series; $_2F_1$ is the two-over-one case, and it is THE second-order ODE solution with three regular singular points at $0,1,\infty$ .

The core insight

A power series whose term ratio is a rational function of the index $n$ is a hypergeometric series; $_2F_1$ is the case with two factors in the numerator and one (besides $n!$ ) in the denominator, and it is THE second-order ODE solution with three regular singular points at $0,1,\infty$ .

Where this lives today

The hypergeometric function is a working tool across modern theoretical physics and computation. In the conformal bootstrap, the four-point conformal blocks of two-dimensional conformal field theory are ₂F₁'s, and the crossing-symmetry equations constraining the space of consistent theories are literally the connection formulae relating $z=0$ , $z=1$ and $z=\infty$ — the monodromy Riemann studied is the theory's fusion and braiding data. In scattering amplitudes, one-loop Feynman integrals evaluated by Mellin–Barnes methods routinely return ₂F₁'s, so its analytic continuation is part of the amplitudes toolkit. In symbolic computation, the Gosper and Zeilberger algorithms for creative telescoping decide hypergeometric summation identities mechanically — Gauss's contiguous relations are now the engine inside a computer-algebra system. Even the plain evaluation of ₂F₁ across the complex plane, needed from lattice QCD to gravitational-wave templates for Kerr black-hole perturbations, remains a live numerical problem handled by the very transformations above.

The physical picture

Start from what you already know. The geometric series $\sum z^n$ has every term equal to the last times $z$ : the ratio is the constant $1$ , and the sum blows up at $|z|=1$ . Now let that ratio depend on where you are in the sum. The hypergeometric ratio is $\frac{(a+n)(b+n)}{(c+n)(n+1)}\,z$ , a rational function of $n$ that still tends to $z$ as $n\to\infty$ — so the radius of convergence is still $1$ — but at finite $n$ it is sculpted by three "resonances": two zeros of the numerator at $n=-a,-b$ and one pole at $n=-c$ . Those numbers $a$ , $b$ , $c$ are the indicial exponents of the three singular points of the underlying differential equation, and the whole global behaviour of the function is fixed by them. That is the geometric heart of the subject. The hypergeometric ODE lives on the Riemann sphere with three marked points, $0$ , $1$ , and $\infty$ . A Möbius transformation can carry any three points anywhere, so fixing them here costs no generality — every three-singular-point problem is a disguised hypergeometric equation. At each marked point the equation admits two local power-law solutions $w\sim(z-z_0)^\rho$ , and the two exponents $\rho$ are the conserved local data. The first figure is the ledger of that data.

Figure

The Riemann scheme of the hypergeometric equation: regular singular points sit only at z = 0, 1, ∞ with exponent pairs {0, 1−c}, {0, c−a−b}, {a, b}, whose six values sum to 1 (the Fuchs relation); the defining series converges inside the unit disk while the branch cut runs from 1 out to ∞

Two levers convert this one function into a whole zoo. Truncation: when $a=-n$ is a negative integer, $(a)_k$ vanishes for $k>n$ and the series collapses to a degree- $n$ polynomial — the assembly line that produces Legendre, Jacobi, Gegenbauer and Chebyshev polynomials, and the reason those families share so many identities. Confluence: rescale $z\to z/b$ and send $b\to\infty$ so the singular point at $z=1$ collides with the one at $\infty$ ; the two regular singularities merge into one irregular singularity, and ₂F₁ degenerates into the confluent function ₁F₁, parent of the Bessel, Laguerre, Hermite and Coulomb functions. The second figure shows a lever at work: the finite-amplitude pendulum, whose period is an elliptic integral and hence a ₂F₁ evaluated at a concrete number.

Figure

Large-amplitude pendulum period straight from ₂F₁: the ratio T/T₀ = (2/π)K(k) = ₂F₁(½, ½; 1; sin²(θ₀/2)) climbs from 1 and diverges as θ₀ → 180° — the marked values are 1.0174 at 30°, 1.1803 at 90°, and 1.7622 at 150°

The core equations

The starting point is the series itself — the Frobenius solution belonging to the exponent $\rho=0$ at $z=0$ , normalized to $1$ there, with coefficients that are ratios of rising factorials.

_2F_1(a,b;c;z)=\sum_{n=0}^{\infty}\frac{(a)_n\,(b)_n}{(c)_n}\,\frac{z^n}{n!} \qquad (1)

where: $(a)_n$ — the Pochhammer symbol (rising factorial), $(a)_n=a(a+1)\cdots(a+n-1)=\Gamma(a+n)/\Gamma(a)$ , with $(a)_0=1$ ; $a,b$ — numerator parameters, entering symmetrically so that $_2F_1(a,b;c;z)=\,_2F_1(b,a;c;z)$ ; $c$ — denominator parameter, forbidden from the values $0,-1,-2,\dots$ ; $z$ — complex argument. Reading it: the subscripts in "₂F₁" count the numerator (two) and denominator (one, besides the automatic $n!$ ) Pochhammer factors; the series is a deformed geometric series converging for $|z|<1$ . Behind the series is the differential equation it solves — a Fuchsian equation whose three regular singular points are visible in its coefficients: $z(1-z)$ vanishes at $0$ and $1$ , and the degree count forces a third singularity at infinity.

z(1-z)\,w''+\bigl[c-(a+b+1)z\bigr]w'-ab\,w=0 \qquad (2)

where: $w(z)$ — any solution; the Frobenius solution with exponent $0$ at the origin is $w=\,_2F_1(a,b;c;z)$ ; $a,b,c$ — the same parameters as the series. Reading it: the exponents at the three singular points are $\{0,1-c\}$ , $\{0,c-a-b\}$ , $\{a,b\}$ , and they sum to $1$ — the Fuchs relation in the first figure. The series only knows the disk $|z|<1$ . To reach the rest of the plane you need an integral, and Euler supplied one: a Beta-type integral analytic in $z$ everywhere off the ray $[1,\infty)$ .

_2F_1(a,b;c;z)=\frac{\Gamma(c)}{\Gamma(b)\,\Gamma(c-b)}\int_0^1 t^{b-1}(1-t)^{c-b-1}(1-zt)^{-a}\,\mathrm{d}t \qquad (3)

where: $\Gamma$ — the Gamma function; the prefactor is exactly $1/B(b,c-b)$ with $B$ the Beta function; the representation is valid for $\mathrm{Re}\,c>\mathrm{Re}\,b>0$ . Reading it: the factor $(1-zt)^{-a}$ is finite for all $t\in[0,1]$ unless $z$ lies on $[1,\infty)$ , so the integral defines the analytic continuation of ₂F₁ to $\mathbb{C}\setminus[1,\infty)$ . At the boundary point $z=1$ the function collapses to a ratio of four Gamma functions — Gauss's summation theorem, the most quoted ₂F₁ identity and the one certifying convergence on the unit circle.

_2F_1(a,b;c;1)=\frac{\Gamma(c)\,\Gamma(c-a-b)}{\Gamma(c-a)\,\Gamma(c-b)},\qquad \mathrm{Re}(c-a-b)>0 \qquad (4)

where: $z=1$ — the boundary singular point; $\mathrm{Re}(c-a-b)>0$ — the convergence condition on $|z|=1$ , from the larger $z=1$ exponent being positive. Reading it: the value at the edge of the disk is fixed entirely by the local exponents, encoded through the Gamma function. The transformation theory lets you move the argument at will. Pfaff's transformation applies a Möbius map that fixes $0$ and swaps $1$ with $\infty$ , converting a value near or beyond the unit circle into one inside it.

_2F_1(a,b;c;z)=(1-z)^{-a}\,_2F_1\!\left(a,\,c-b;\,c;\,\frac{z}{z-1}\right) \qquad (5)

where: $z/(z-1)$ — the Möbius image of $z$ ; for real $z<0$ this maps into $(0,1)$ , accelerating convergence. Reading it: a companion connection formula (Kummer's) expresses the same function in the basis natural at $z=\infty$ , using powers $(-z)^{-a}$ and $(-z)^{-b}$ fixed by the exponents there — the machinery for large $|z|$ . Applying Pfaff twice yields Euler's transformation, trading $(a,b)$ for $(c-a,c-b)$ at the price of a prefactor carrying the second $z=1$ exponent.

_2F_1(a,b;c;z)=(1-z)^{c-a-b}\,_2F_1(c-a,\,c-b;\,c;\,z) \qquad (6)

where: $c-a-b$ — the nontrivial exponent at $z=1$ ; the identity holds throughout the cut plane. Reading it: the prefactor $(1-z)^{c-a-b}$ carries exactly the $z=1$ singular behaviour, leaving a ₂F₁ regular there. Differentiation walks along the parameter lattice: the derivative of a ₂F₁ is another ₂F₁ with every parameter raised by one, immediate from differentiating the series term by term.

\frac{\mathrm{d}}{\mathrm{d}z}\,_2F_1(a,b;c;z)=\frac{ab}{c}\,_2F_1(a+1,\,b+1;\,c+1;\,z) \qquad (7)

where: $a+1,b+1,c+1$ — the raised parameters; the prefactor $ab/c$ is the first Taylor coefficient. Reading it: derivatives never leave the hypergeometric family, which is why ₂F₁ closes under the operations of solving differential equations. Finally, Gauss found fifteen contiguous relations — three-term identities with rational coefficients connecting a ₂F₁ to two neighbours on the integer parameter lattice. One representative:

c\,_2F_1(a-1,b;c;z)-c\,_2F_1(a,b;c;z)+bz\,_2F_1(a,b+1;c+1;z)=0 \qquad (8)

where: $a\pm1,\,b\pm1,\,c\pm1$ — the neighbouring parameter shifts; the coefficients are rational in $a,b,c,z$ . Reading it: across the parameter lattice ₂F₁ satisfies three-term recurrences — the basis of the classical polynomial recurrences and of stable numerical evaluation.

Derivation AThe Frobenius series solves the hypergeometric equation

The goal is to prove that the defining series is genuinely a solution of the hypergeometric ODE, and in doing so to derive the Pochhammer coefficients rather than assume them. We substitute an undetermined power series into the equation, collect the coefficient of a generic power $z^n$ , and discover that the recursion factorizes so cleanly that its solution is forced to be the ratio of rising factorials $(a)_n(b)_n/[(c)_n\,n!]$ . The crux is a single algebraic identity: the quadratic in $n$ produced by the equation is exactly $(n+a)(n+b)$ , and recognizing that factorization is the whole game.

Step 1. Seek $w=\sum_{n\ge0}u_n z^n$ with $u_0=1$ and differentiate term by term: $w'=\sum_{n\ge1}n\,u_n z^{n-1}$ and $w''=\sum_{n\ge2}n(n-1)u_n z^{n-2}$ . Why: term-by-term differentiation is legitimate inside the radius of convergence, and starting from $u_0=1$ picks out the Frobenius solution belonging to exponent $\rho=0$ at the origin.

Step 2. Insert into $z(1-z)w''+[c-(a+b+1)z]w'-ab\,w=0$ and gather the coefficient of $z^n$ . The pieces $z\,w''$ and $c\,w'$ contribute terms in $u_{n+1}$ , while $-z^2 w''$ , $-(a+b+1)z\,w'$ and $-ab\,w$ contribute terms in $u_n$ , giving $(n+1)n\,u_{n+1}+c(n+1)u_{n+1}-[n(n-1)+(a+b+1)n+ab]u_n=0$ . Why: the shift by one power between the $u_{n+1}$ and $u_n$ groups is what turns the ODE into a two-term recurrence rather than a messier relation.

Step 3. Factor the bracket. Expanding, $n(n-1)+(a+b+1)n+ab=n^2+(a+b)n+ab=(n+a)(n+b)$ , so $u_{n+1}=\dfrac{(n+a)(n+b)}{(n+1)(n+c)}\,u_n$ . Why: this factorization is the heart of the theory — the numerator roots $-a,-b$ and the denominator root $-c$ reappear as indicial exponents at $z=\infty$ and $z=0$ .

Step 4. Telescope from $u_0=1$ : multiplying the ratios gives $u_n=\dfrac{(a)_n(b)_n}{(c)_n\,n!}$ , exactly the coefficient in equation (1). As a check, $u_2=\dfrac{a(a+1)\,b(b+1)}{2\,c(c+1)}$ by both routes. Why: recovering the defining coefficients confirms the series solves the ODE, and the $z^2$ cross-check guards against an algebra slip. *As a cross-check, the recursion $u_{n+1}=(n+a)(n+b)/[(n+1)(n+c)]\,u_n$ and the closed form $u_n=(a)_n(b)_n/[(c)_n n!]$ agree through $n=2$ .*

Go deeper · Derivation B · 1 minEuler's integral reproduces the series and continues it past the disk

Derivation BEuler's integral reproduces the series and continues it past the disk

The aim is to establish Euler's integral representation and, with it, the analytic continuation that carries ₂F₁ beyond the unit disk. The strategy is to expand the only $z$ -dependent factor as a binomial series, integrate the resulting powers of $t$ against the Beta kernel term by term, and watch the Beta integral manufacture exactly the $(b)_n/(c)_n$ ratios the definition demands. Because the integral converges for every $z$ off the ray $[1,\infty)$ while the series converges only for $|z|<1$ , the identity does double duty: it verifies the representation and it is the continuation.

Step 1. Expand the argument-dependent factor binomially, $(1-zt)^{-a}=\sum_{n\ge0}\frac{(a)_n}{n!}(zt)^n$ , which converges uniformly for $|zt|\le|z|<1$ . Why: uniform convergence on $t\in[0,1]$ is what licenses swapping the sum and the integral in the next step.

Step 2. Integrate each power of $t$ against the kernel: $\int_0^1 t^{b+n-1}(1-t)^{c-b-1}\,\mathrm{d}t=B(b+n,c-b)=\dfrac{\Gamma(b+n)\Gamma(c-b)}{\Gamma(c+n)}$ . Why: this is the exact point where the Beta function enters the theory of ₂F₁ — the numerator parameter $b$ and denominator parameter $c$ are the two Beta arguments.

Step 3. Assemble the prefactor with the Beta value: $\dfrac{\Gamma(c)}{\Gamma(b)\Gamma(c-b)}\cdot\dfrac{\Gamma(b+n)\Gamma(c-b)}{\Gamma(c+n)}=\dfrac{\Gamma(c)\Gamma(b+n)}{\Gamma(b)\Gamma(c+n)}=\dfrac{(b)_n}{(c)_n}$ , using $\Gamma(b+n)/\Gamma(b)=(b)_n$ and $\Gamma(c)/\Gamma(c+n)=1/(c)_n$ . Why: the $\Gamma(c-b)$ cancels cleanly, leaving precisely the ratio of Pochhammer symbols that the series requires.

Step 4. Sum over $n$ : $\;\dfrac{\Gamma(c)}{\Gamma(b)\Gamma(c-b)}\int_0^1 t^{b-1}(1-t)^{c-b-1}(1-zt)^{-a}\,\mathrm{d}t=\sum_{n\ge0}\dfrac{(a)_n(b)_n}{(c)_n}\dfrac{z^n}{n!}=\,_2F_1(a,b;c;z)$ . The right side is the series, but the left side converges and is analytic for all $z\notin[1,\infty)$ . Why: the two sides agree on the overlap $|z|<1$ and the integral is analytic on a larger domain, so by the identity theorem the integral is the unique analytic continuation. *As a consistency check, setting $z=1$ reduces the integral to $B(b,c-a-b)/B(b,c-b)$ times the prefactor, reproducing Gauss's summation theorem.*

Go deeper · Derivation C · 1 minGauss's summation theorem from the integral at z = 1

Derivation CGauss's summation theorem from the integral at z = 1

The goal is to evaluate ₂F₁ in closed form at its boundary singular point $z=1$ , obtaining Gauss's ratio of four Gamma functions and seeing exactly why the convergence condition is $\mathrm{Re}(c-a-b)>0$ . We take $z\to1$ directly inside Euler's integral: the factor $(1-zt)^{-a}$ becomes $(1-t)^{-a}$ , merging with the existing power of $(1-t)$ to leave a single pure Beta integral. Whether that integral converges at the endpoint $t=1$ is precisely the condition $\mathrm{Re}(c-a-b)>0$ , so the value and its domain of validity come out together.

Step 1. Set $z=1$ in the kernel: $(1-zt)^{-a}\big|_{z=1}=(1-t)^{-a}$ , so the integrand becomes $t^{b-1}(1-t)^{c-b-1}(1-t)^{-a}=t^{b-1}(1-t)^{c-a-b-1}$ . Why: the two powers of $(1-t)$ combine, collapsing the argument-dependent representation into a parameter-only Beta integral.

Step 2. Recognize the Beta integral: $\int_0^1 t^{b-1}(1-t)^{c-a-b-1}\,\mathrm{d}t=B(b,c-a-b)=\dfrac{\Gamma(b)\Gamma(c-a-b)}{\Gamma(c-a)}$ , converging at $t=1$ iff $c-a-b-1>-1$ , i.e. $\mathrm{Re}(c-a-b)>0$ . Why: the endpoint behaviour of the integrand is the origin of the convergence condition — nothing is imposed by hand.

Step 3. Multiply by the prefactor and cancel: $\dfrac{\Gamma(c)}{\Gamma(b)\Gamma(c-b)}\cdot\dfrac{\Gamma(b)\Gamma(c-a-b)}{\Gamma(c-a)}=\dfrac{\Gamma(c)\,\Gamma(c-a-b)}{\Gamma(c-a)\,\Gamma(c-b)}$ . Why: the $\Gamma(b)$ cancels, leaving the symmetric four-Gamma form invariant under $a\leftrightarrow b$ .

Step 4. Sanity-check the result. For $a=-1$ the series truncates to $1-b/c=(c-b)/c$ , and the formula gives $\Gamma(c)\Gamma(c+1-b)/[\Gamma(c+1)\Gamma(c-b)]=(c-b)/c$ — they agree. For $a=b=\tfrac12,\,c=1$ the formula has $\Gamma(c-a-b)=\Gamma(0)$ , which diverges, correctly signalling that the complete elliptic integral $K(k)\to\infty$ as $k\to1$ . Why: one finite check and one divergent check together confirm both the value and the boundary of validity. *As a cross-check: $a=-1$ gives $(c-b)/c$ on both sides, and $a=b=\tfrac12,c=1$ diverges as it must for $K(k\to1)$ .*

Limiting cases

Special values of the parameters collapse ₂F₁ onto functions with their own names. Each row below is an identity you can verify by comparing Taylor coefficients.

Case	Result	Meaning
Geometric / binomial	$_2F_1(a,b;b;z)=(1-z)^{-a}$	cancelling $c=b$ leaves $\sum(a)_n z^n/n!$ ; the choice $a=1$ is the plain geometric series $\sum z^n$
Logarithm	$z\,_2F_1(1,1;2;z)=-\ln(1-z)$	here $(1)_n=n!$ and $(2)_n=(n+1)!$ , giving $\sum z^{n+1}/(n+1)$
Arcsine	$z\,_2F_1(\tfrac12,\tfrac12;\tfrac32;z^2)=\arcsin z$	a quadratic-argument case matching the Taylor series of $\arcsin$
Complete elliptic integral	$K(k)=\tfrac{\pi}{2}\,_2F_1(\tfrac12,\tfrac12;1;k^2)$	the pendulum and arithmetic–geometric mean live here; also $E(k)=\tfrac\pi2\,_2F_1(-\tfrac12,\tfrac12;1;k^2)$
Legendre polynomial	$P_n(x)=\,_2F_1(-n,n+1;1;\tfrac{1-x}{2})$	the negative-integer first parameter truncates the series to degree $n$
Jacobi polynomial	$P_n^{(\alpha,\beta)}(x)\propto\,_2F_1(-n,n+\alpha+\beta+1;\alpha+1;\tfrac{1-x}{2})$	the general classical orthogonal polynomial, containing all the others

Common trapThe series is the function everywhere.

A common error is to treat the power series and the function ₂F₁ as the same object. They are not. The series diverges for $|z|>1$ — but the function is well defined there, by Euler's integral and the Pfaff, Euler and Kummer transformations, which give convergent representations covering the whole cut plane. Quoting "₂F₁ diverges for $|z|>1$ " confuses the representation with the object. A second, independent trap concerns $c$ : the series has poles at $c=0,-1,-2,\dots$ because a factor $(c)_n$ vanishes. When $c$ is a non-positive integer, switch to the regularized function $\mathbf{F}(a,b;c;z)=\,_2F_1(a,b;c;z)/\Gamma(c)$ , entire in all of $a,b,c$ — the pole of $1/\Gamma(c)$ exactly cancels the divergence.

Worked examples

1 · The finite-amplitude pendulum (fully worked)

A pendulum of length $L$ is released from rest at amplitude $\theta_0=90^\circ$ . Its exact period is $T=4\sqrt{L/g}\,K(k)$ with $k=\sin(\theta_0/2)$ , and $K(k)=\tfrac{\pi}{2}\,_2F_1(\tfrac12,\tfrac12;1;k^2)$ . By what factor does the true period exceed the small-angle value $T_0=2\pi\sqrt{L/g}$ ?

Step 1. Fix the modulus. At $\theta_0=90^\circ$ the half-angle is $45^\circ$ , so $k=\sin45^\circ=1/\sqrt2$ and $k^2=\tfrac12=0.5$ . This is safely inside the disk $|z|<1$ , so the defining series converges.

Step 2. Sum $_2F_1(\tfrac12,\tfrac12;1;0.5)$ with the term ratio $\dfrac{u_{n+1}}{u_n}=\dfrac{(n+\tfrac12)^2}{(n+1)^2}(0.5)$ . The terms are $u_0=1$ , $u_1=0.125$ , $u_2=0.03516$ , $u_3=0.01221$ , $u_4=0.004673$ , $u_5=0.001893$ , and the tail adds $\approx0.00173$ , giving $_2F_1(\tfrac12,\tfrac12;1;0.5)=\mathbf{1.18034}$ .

Step 3. Convert to the period ratio. Since $T/T_0=\dfrac{4\sqrt{L/g}\,K(k)}{2\pi\sqrt{L/g}}=\dfrac{2K(k)}{\pi}=\,_2F_1(\tfrac12,\tfrac12;1;k^2)$ — the $\pi/2$ factors cancel — the answer is the hypergeometric value: $T/T_0=\mathbf{1.1803}$ . A $90^\circ$ pendulum swings about 18% slower than the small-angle formula predicts.

Step 4. Read the physics off the second figure. As $\theta_0\to180^\circ$ the argument $k^2\to1$ and $_2F_1(\tfrac12,\tfrac12;1;1)$ diverges because $c-a-b=0$ fails Gauss's condition — the period grows without bound, exactly as a pendulum balanced at the inverted position takes forever to fall. T/T₀ = 1.1803 at 90°, diverging logarithmically as θ₀ → 180°.

2 · Truncation builds a Legendre polynomial

Show directly that setting $a=-2$ in the Legendre representation reproduces $P_2(x)=\tfrac12(3x^2-1)$ , and confirm the general rule that $a=-n$ gives a degree- $n$ polynomial.

Step 1. Write the representation with $n=2$ : $P_2(x)=\,_2F_1(-2,3;1;u)$ , $u=\tfrac{1-x}{2}$ . Because $a=-2$ , the factor $(-2)_k$ is $1,-2,2,0,0,\dots$ , so only $k=0,1,2$ survive — the series is automatically a quadratic in $u$ . The truncation is the whole point: a negative-integer parameter kills every term past $k=n$ .

Step 2. Compute the three surviving terms. With $(-2)_0(3)_0/[(1)_0 0!]=1$ , $\;(-2)_1(3)_1/[(1)_1 1!]=(-2)(3)/1=-6$ , and $(-2)_2(3)_2/[(1)_2 2!]=(2)(12)/(2\cdot2)=6$ , the polynomial is $_2F_1(-2,3;1;u)=1-6u+6u^2$ .

Step 3. Substitute $u=\tfrac{1-x}{2}$ : $\;1-6\cdot\tfrac{1-x}{2}+6\cdot\tfrac{(1-x)^2}{4}=1-3(1-x)+\tfrac32(1-x)^2$ . Expanding, $=1-3+3x+\tfrac32(1-2x+x^2)=-2+3x+\tfrac32-3x+\tfrac32x^2=\tfrac32x^2-\tfrac12=\mathbf{\tfrac12(3x^2-1)}$ , which is precisely $P_2(x)$ .

Step 4. Generalize. For any non-negative integer $n$ , $a=-n$ makes $(a)_k=0$ for $k>n$ , so $_2F_1(-n,n+1;1;\tfrac{1-x}{2})$ is a degree- $n$ polynomial — the single mechanism producing the entire Legendre family, and with a shifted third parameter the Gegenbauer, Chebyshev and Jacobi families too. A negative-integer numerator parameter turns ₂F₁ into a degree-n orthogonal polynomial.

Physics of the portrait

The Gaussian hypergeometric function carries Gauss's name for good reason: he was the first to treat the series as a function in its own right, to find its value at $z=1$ , and to organize its parameter shifts into a closed algebraic system. But the story runs from Wallis through Euler to Gauss, Kummer and Riemann, each of whom saw a different face of the same object.

Origins · 1 minCarl Friedrich Gauss and how this was discovered

Carl Friedrich Gauss

1777–1855 · the hypergeometric series, its summation theorem, and the contiguous relations

By the time Gauss took up the hypergeometric series it already had a prehistory. John Wallis had used the word "hypergeometric" in his Arithmetica Infinitorum of 1655, and Leonhard Euler had written down both the series and an integral for it and found the transformation that now bears his name. What Gauss did, in his 1812 memoir Disquisitiones generales circa seriem infinitam, was to make the series an object of systematic study: he introduced clean notation, proved the summation theorem $_2F_1(a,b;c;1)=\Gamma(c)\Gamma(c-a-b)/[\Gamma(c-a)\Gamma(c-b)]$ , established rigorous convergence criteria on the unit circle at a time when convergence was rarely treated with care, and derived the fifteen contiguous relations linking neighbouring parameter values. His motivation was partly geodetic — elliptic integrals and the arithmetic–geometric mean, which he had studied since his teens, are hypergeometric — and partly fascination with what one function could contain. The theory was completed by others: Ernst Kummer in 1836 catalogued the twenty-four solutions and the quadratic transformations, and Bernhard Riemann in 1857 stripped the subject to its essence, showing that everything follows from the exponents at the three singular points and the monodromy of the solutions around them — the viewpoint that turned a series into geometry.

History

1655

— Wallis uses the term "hypergeometric" in his Arithmetica Infinitorum

1748

— Euler gives the series and an integral representation for it

1812

— Gauss proves the summation theorem and the fifteen contiguous relations

1836

— Kummer classifies the twenty-four solutions and the quadratic transformations

1857

— Riemann recasts the equation through the P-symbol and monodromy

Going further

Texts. Stone & Goldbart, Mathematics for Physics (Cambridge, 2009), Ch. 16 · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 18 · Hassani, Mathematical Physics, 2nd ed., Ch. 15 · Whittaker & Watson, A Course of Modern Analysis, 4th ed., Ch. 14 · Abramowitz & Stegun, Ch. 15 · NIST DLMF, Ch. 15. Key self-consistency checks: the ODE bracket factors as $(n+a)(n+b)$ ; integrating $(1-zt)^{-a}$ against the Beta kernel recovers the series; Gauss's formula gives $(c-b)/c$ at $a=-1$ and diverges at $a=b=\tfrac12,c=1$ ; the special cases $(1-z)^{-a}$ , $-\ln(1-z)$ , $\arcsin z$ and $P_2(x)=\tfrac12(3x^2-1)$ are confirmed by Taylor coefficients; and $_2F_1(\tfrac12,\tfrac12;1;k^2)=1.0174,\,1.1803,\,1.7622$ at $\theta_0=30^\circ,90^\circ,150^\circ$ .

Builds on: Gamma Function · Beta Function · Analytic Continuation and Poles of the Gamma Function. Leads to: Legendre Polynomials · Bessel Functions of the First and Second Kind · Orthogonal Polynomials: General Theory · Recurrence Relations for Special Functions.

Ask the tutor: Ask me to show how any one function you know — a logarithm, an elliptic integral, a Legendre polynomial, or a Bessel function — falls out of ₂F₁ by choosing its parameters.

Why is it called "hypergeometric" — what is beyond geometric about it?

The name goes back to Wallis and marks a hierarchy of series classified by their term ratio. A geometric series has a constant ratio, $u_{n+1}/u_n=r$ ; a hypergeometric series relaxes this to a rational function of the index, $u_{n+1}/u_n=R(n)$ . The Gaussian ₂F₁ is the simplest genuinely new case, with $R(n)=\frac{(n+a)(n+b)}{(n+c)(n+1)}z$ . Because that ratio still tends to $z$ as $n\to\infty$ , the series shares the geometric series' unit radius of convergence, but its finite- $n$ shaping by $a,b,c$ is what lets it encode differential equations and reproduce named functions. "Hyper" simply means "one level above" the geometric series.

Why exactly three singular points — what is special about the number three?

Three is the number of points you can fix on the Riemann sphere with a Möbius transformation, whose group has exactly three complex degrees of freedom. An equation with fewer than three regular singular points is too rigid to be interesting (two give power laws, one gives constants), while an equation with three has, after normalizing their locations, only the exponents left as data. This is why Riemann could prove that every three-singular-point Fuchsian equation is hypergeometric: the geometry has no room for anything else. Equations with four regular singular points (the Heun equation) are genuinely harder and are not reducible to ₂F₁ in general.

How do you actually compute ₂F₁ when |z| is not small?

You never sum the raw series outside $|z|<1$ , and even inside it you avoid $z$ near the boundary where convergence is slow. Instead the transformations act as a routing table. If $|z|$ is near or above $1$ , Pfaff's transformation maps the argument to $z/(z-1)$ , which for real $z<0$ lands in $(0,1)$ . If $z$ is near $1$ , a connection formula re-expands the function in powers of $1-z$ . If $|z|$ is large, Kummer's formula re-expresses ₂F₁ in powers of $1/z$ . Production libraries partition the plane into regions, apply whichever transformation converges fastest, then refine with the contiguous relations — the mathematics of the core-equations section is, quite literally, the numerical algorithm.

Why do all the classical orthogonal polynomials come from ₂F₁?

Because they all solve second-order equations of hypergeometric type. Legendre's, Jacobi's, Gegenbauer's and Chebyshev's equations are each the hypergeometric equation (or its confluent limit) with parameters chosen so that one numerator parameter is a negative integer $-n$ . That single choice truncates the series to a degree- $n$ polynomial, and the shared parentage explains why these families have parallel Rodrigues formulae, three-term recurrences and generating functions — inherited from ₂F₁'s derivative formula and contiguous relations. Hermite and Laguerre polynomials arise the same way from the confluent ₁F₁. In short, "classical orthogonal polynomial" is essentially a synonym for "truncated hypergeometric function with a weight."

What are the contiguous relations good for?

Gauss's fifteen contiguous relations connect a ₂F₁ to two neighbours whose parameters differ by $\pm1$ . Theoretically, iterating them proves the three-term recurrences of every orthogonal polynomial family and the ladder relations of wavefunctions built from ₁F₁. Computationally, they let you fill a whole lattice of ₂F₁ values from a couple of seed evaluations by stable recursion, far cheaper than resumming the series each time. And in creative telescoping, Zeilberger's algorithm treats them as a certificate that mechanically proves hypergeometric summation identities — what Gauss derived by hand is now the backbone of automated identity-proving.

When must I use the regularized function F instead of ₂F₁?

Whenever the denominator parameter $c$ is zero or a negative integer. The ordinary series has factors $(c)_n$ downstairs, and once $c\in\{0,-1,-2,\dots\}$ one of those vanishes and the term blows up — ₂F₁ has a pole there. The regularized function $\mathbf{F}(a,b;c;z)=\,_2F_1(a,b;c;z)/\Gamma(c)$ cures this: since $1/\Gamma(c)$ has a zero exactly where ₂F₁ has its pole, the product is finite and in fact entire in all of $a,b,c$ . You meet this when a physical limit drives $c$ toward a non-positive integer — degenerate Coulomb or Legendre cases — where the regularized function keeps the formulae well behaved. For generic $c$ the two differ only by the constant $\Gamma(c)$ .