Confluent Hypergeometric Functions

The confluent hypergeometric functions are the master key to the special functions of physics. Behind Bessel functions, Coulomb wavefunctions, Laguerre and Hermite polynomials, parabolic-cylinder functions, the error function and the incomplete gamma function stands a single ordinary differential equation with exactly one regular singular point and one irregular singular point of rank one — Kummer's equation. Learn the analytic structure of its two solutions once, and the asymptotics, connection formulae and Stokes phenomena of that entire menagerie follow as corollaries rather than as separate theories to be memorised. The name is a piece of geometry made into a limit. The Gauss equation for ₂F₁ carries three regular singular points, at 0, 1 and ∞. Rescale the independent variable and let one parameter run to infinity, and the singularity at 1 slides out along the real axis until it collides with the one already at infinity. Two regular singular points "flow together" — confluunt — into a single irregular one, and the surviving equation is Kummer's. This coalescence is why the confluent functions inherit an exponential, e^z, in their large-argument behaviour: an essential singularity is the fused remnant of two ordinary power-law singularities. The two-dimensional solution space is spanned by complementary characters. Kummer's M(a,b;z) is entire, fixed by regularity at the origin, and generically explodes like e^z far away. Tricomi's U(a,b;z) is the unique solution that decays algebraically at +∞ and generically branches at the origin. Physical eigenvalue problems demand one function be acceptable at both ends at once — a clash that is impossible unless the series terminates, and the terminating condition a = −n is exactly quantization.

Key takeaways

Kummer's equation z w″ + (b − z) w′ − a w = 0 has a regular singular point at z = 0 and an irregular one at z = ∞; its two solutions are the entire Kummer function M(a,b;z) = ₁F₁(a;b;z) and Tricomi's U(a,b;z).
They are the confluent limit of the Gauss function: merging two of the three regular singular points of ₂F₁ sends z = 1 out to z = ∞, leaving one regular and one irregular singularity.
Integral representations tie them to the Gamma and Beta functions: M is a beta integral with an inserted exponential, U a Laplace-type integral over the half-line.
As z → +∞, M generically grows like Γ(b)/Γ(a) · e^z z^{a−b} while U decays like z^{−a}; the relative weight of the two behaviours jumps across Stokes lines.
Terminating the series at a = −n produces Laguerre and Hermite polynomials — the bound states of the Coulomb and harmonic-oscillator problems, with the hydrogen spectrum E_n = −1/(2n²) falling straight out.
Key idea: collapse the Gauss equation's three singular points to two by a confluent limit; the resulting Kummer equation has the entire solution M (regular at 0, exponentially growing at ∞) and the Tricomi solution U (singular at 0 in general, algebraically decaying at ∞), and every special function with one regular and one irregular singular point is one of these in disguise.

The core insight

Collapse the Gauss equation's three singular points to two by a confluent limit; the resulting Kummer equation has the entire solution M (regular at 0, exponentially growing at ∞) and the Tricomi solution U (singular at 0 in general, algebraically decaying at ∞). Every special function with one regular and one irregular singular point is one of these in disguise.

Where this lives today

The confluent functions are not a historical curiosity; they are live infrastructure. Every scientific computing library ships them — SciPy's hyp1f1 and hyperu, the GNU Scientific Library, Boost — because Coulomb wavefunctions F_ℓ and G_ℓ are needed the moment a charged particle scatters. In nuclear astrophysics the Gamow factor |Γ(ℓ+1+iη)|², an analytic continuation of the same Γ functions that build M and U, sets the sub-barrier fusion rates and astrophysical S-factors that govern how stars burn and how the elements were made. In atomic and molecular physics the parabolic-cylinder functions — U-type confluent functions — give the exact Landau–Zener transition probability, the workhorse formula for how fast you may sweep a control field across an avoided crossing in a superconducting qubit before you lose adiabaticity. And the confluent equation is a favourite laboratory for exponential asymptotics: the Stokes phenomenon it exhibits in its cleanest form — the switching-on of a subdominant exponential across a ray — is exactly the object modern resurgence theory turns into a precise, computable transseries.

The physical picture

The story of the confluent functions is the story of two singular points becoming one. A regular singular point of a second-order linear ODE contributes power-law solutions z^ρ set by an indicial equation; an irregular singular point contributes an essential singularity, an exponential. The confluent equation is the simplest place where both appear together, and the first figure shows how it is manufactured out of the fully-regular Gauss equation by sliding one singularity into another.

Figure

The confluence of singular points. The Gauss equation for ₂F₁ has three regular singular points at 0, 1 and ∞. Rescaling x = z/β pushes the point at x = 1 out to z = β, and as β → ∞ it merges with the point at infinity into one irregular singular point of rank 1 — the surviving Kummer equation z w″ + (b − z) w′ − a w = 0

Once the equation exists, its two solutions divide the labour. Selecting the solution that stays finite at the origin picks out Kummer's M, an everywhere-convergent power series that, having no reason to be small far away, carries the full weight of the merged singularity's exponential and blows up like e^z. Selecting instead the solution that decays at infinity picks out Tricomi's U, which pays for that good behaviour with a branch point at the origin. The second figure makes the split concrete on the one case where both functions are elementary, so every number can be checked by hand.

Figure

The two solutions of Kummer's equation for a = 1, b = 2, where M(1,2;z) = (e^z − 1)/z and U(1,2;z) = 1/z exactly. The entire solution M blows up like e^z/z, reaching 13.40 at z = 4, while the recessive solution U decays algebraically like z^−a = 1/z, equal to 0.250 at z = 4

The core equations

Everything begins with the equation itself. Kummer's equation is the second-order linear ODE whose coefficient of the highest derivative vanishes linearly at the origin (making z = 0 a regular singular point) and whose remaining coefficient grows linearly (making z = ∞ irregular of rank one) — the canonical two-singularity equation.

z\,\frac{\mathrm{d}^2 w}{\mathrm{d}z^2}+(b-z)\,\frac{\mathrm{d}w}{\mathrm{d}z}-a\,w=0 \qquad (1)

where: $a$ — first parameter; controls the asymptotic power $z^{-a}$ and the termination condition $a=-n$ ; $b$ — second parameter; the indicial exponents at $z=0$ are $0$ and $1-b$ ; $z$ — complex independent variable, with $z=0$ regular singular and $z=\infty$ irregular singular of rank one; $w$ — any solution. Reading: two parameters $a,b$ tune a single universal shape, and their values decide which named special function you see. The solution regular at the origin is Kummer's function, built by the Frobenius method as an everywhere-convergent power series. Its coefficients are ratios of Pochhammer symbols, which is why it is written as the ₁F₁ hypergeometric series.

M(a,b;z)={}_1F_1(a;b;z)=\sum_{n=0}^{\infty}\frac{(a)_n}{(b)_n}\,\frac{z^n}{n!} \qquad (2)

where: $(a)_n=a(a+1)\cdots(a+n-1)=\Gamma(a+n)/\Gamma(a)$ is the rising factorial (Pochhammer symbol), with $(a)_0=1$ ; $(b)_n$ is the same for $b$ and requires $b\notin\{0,-1,-2,\dots\}$ so no denominator vanishes. Reading: the ratio of successive terms is $(a+n)z/[(b+n)(n+1)]\to 0$ , so the series converges for every finite $z$ — $M$ is entire. The series has a companion integral representation that ties $M$ directly to the Beta function: it is a beta integral with an exponential $e^{zt}$ folded in. This form is the springboard for the asymptotics and for many identities.

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

where: $\Gamma$ — Euler's Gamma function; the representation is valid for $\operatorname{Re}\,b>\operatorname{Re}\,a>0$ so the endpoint singularities are integrable. Reading: expanding $e^{zt}$ and integrating term by term reproduces the series (2), because each term is a Beta integral $B(a+n,b-a)$ . The solution selected instead by decay at infinity is Tricomi's function. Its natural representation is a Laplace-type integral over the half-line, which manifestly decays as $\operatorname{Re}\,z\to+\infty$ .

U(a,b;z)=\frac{1}{\Gamma(a)}\int_0^\infty e^{-zt}\,t^{a-1}(1+t)^{b-a-1}\,\mathrm{d}t \qquad (4)

where: the integral converges for $\operatorname{Re}\,z>0$ and $\operatorname{Re}\,a>0$ ; elsewhere $U$ is fixed by analytic continuation. Reading: because the factor $(1+t)^{b-a-1}$ has no zero on $[0,\infty)$ , this integral has no exponentially growing part — $U$ is the unique recessive solution at $+\infty$ . $U$ is a definite linear combination of the two Frobenius solutions built at $z=0$ . The connection coefficients are ratios of Gamma functions, and the second piece carries the branch $z^{1-b}$ that makes $U$ generically singular at the origin.

U(a,b;z)=\frac{\Gamma(1-b)}{\Gamma(a-b+1)}\,M(a,b;z)+\frac{\Gamma(b-1)}{\Gamma(a)}\,z^{1-b}\,M(a-b+1,2-b;z) \qquad (5)

where: $z^{1-b}$ — the second indicial branch, singular at $z=0$ when $\operatorname{Re}\,b>1$ ; the formula requires $b\notin\mathbb{Z}$ and is completed for integer $b$ by a limiting (logarithmic) procedure. Reading: this line bridges the small- $z$ world of $M$ and the large- $z$ world of $U$ . The behaviour at the irregular point is the heart of the theory. As $|z|\to\infty$ two independent behaviours compete: a dominant exponential and a subdominant algebraic tail. The entire function $M$ contains a fixed amount of each.

M(a,b;z)\sim\frac{\Gamma(b)}{\Gamma(a)}\,e^{z}z^{a-b}+\frac{\Gamma(b)}{\Gamma(b-a)}(-z)^{-a},\qquad U(a,b;z)\sim z^{-a} \qquad (6)

where: $e^{z}z^{a-b}$ is the dominant (recessive on other rays) exponential piece — the signature of the irregular singularity; $(-z)^{-a}$ is the subdominant algebraic piece, whose relative weight jumps as $z$ crosses a Stokes line. Reading: $M$ blows up because the coefficient $\Gamma(b)/\Gamma(a)$ of its exponential is generically nonzero; $U$ is precisely the combination in which that coefficient is zero, leaving only $z^{-a}$ . Finally, a pair of workhorse identities. Kummer's transformation extracts the exponential and so trades growth for decay, and the contiguous relation links functions with neighbouring parameters — the basis of stable numerical evaluation.

M(a,b;z)=e^{z}\,M(b-a,b;-z),\qquad (b-a)\,M(a-1,b;z)+(2a-b+z)\,M(a,b;z)-a\,M(a+1,b;z)=0 \qquad (7)

where: the first identity (Kummer's first transformation) maps $M(a,b;z)$ , growing like $e^z$ , onto $e^z$ times a function of $-z$ ; the second is the three-term recurrence in $a$ . Reading: the transformation is a symmetry under $a\mapsto b-a$ , $z\mapsto -z$ ; the recurrence climbs the ladder of $a$ -values from two seeds.

Derivation AThe confluent limit of the Gauss hypergeometric function

The goal is to earn the name "confluent" by constructing Kummer's equation and Kummer's function as a genuine limit of the Gauss hypergeometric equation, watching two of its three regular singular points slide together into one irregular point. We rescale the variable so one singularity is dragged to infinity, take the limit of the differential equation, then confirm the ₂F₁ series collapses term by term onto the ₁F₁ series — so the identity holds for both the equation and its distinguished solution.

Step 1. Write the Gauss hypergeometric equation for w(x) = ₂F₁(a, β; b; x), whose three regular singular points sit at x = 0, 1, ∞:

x(1-x)\,w''+\big[b-(a+\beta+1)x\big]w'-a\beta\,w=0

Why: this is the parent equation, and the whole construction is a controlled statement about what happens to its singular point at x = 1.

Step 2. Rescale the variable so that the parameter β sets a shrinking scale, x = z/β, under which d/dx = β d/dz. The singular point x = 1 is now located at z = β:

x=\frac{z}{\beta},\qquad \frac{\mathrm{d}}{\mathrm{d}x}=\beta\,\frac{\mathrm{d}}{\mathrm{d}z}

Why: fixing z while sending β → ∞ freezes the point z = 0 in place but launches the point at x = 1 off to z = β = ∞, which is exactly the coalescence we want to engineer.

Step 3. Substitute and divide through by one power of β. With primes now meaning d/dz,

\frac{z}{\beta}\Big(1-\frac{z}{\beta}\Big)\beta\,w''+\Big[b-(a+\beta+1)\frac{z}{\beta}\Big]w'-a\,w=0

Why: dividing by β is what keeps the equation finite in the limit — the a β term that would blow up is exactly balanced by the β hidden in the second-derivative coefficient.

Step 4. Take β → ∞. The term z²/β vanishes, the coefficient (a + β + 1)/β → 1, and the equation contracts to

z\,w''+(b-z)\,w'-a\,w=0

Why: two regular singular points have fused into a single irregular one at infinity, and what survives is precisely Kummer's equation (1).

Step 5. Check the same limit on the series. Since (β)ₙ/βⁿ = ∏(1 + k/β) → 1, each Gauss term collapses onto a Kummer term,

\frac{(a)_n(\beta)_n}{(b)_n\,n!}\Big(\frac{z}{\beta}\Big)^n=\frac{(a)_n}{(b)_n}\frac{z^n}{n!}\cdot\frac{(\beta)_n}{\beta^n}\;\longrightarrow\;\frac{(a)_n}{(b)_n}\frac{z^n}{n!}

Why: this confirms that the distinguished regular solution ₂F₁ tends to M term by term, so lim(β→∞) ₂F₁(a, β; b; z/β) = M(a, b; z) — the confluence holds for the solution, not merely the equation.

Go deeper · Derivation B · 1 minLeading asymptotics of M from the Kummer integral

Derivation BLeading asymptotics of M from the Kummer integral

The goal is to extract, rigorously, the exponential growth M(a, b; z) ~ Γ(b)/Γ(a) · e^z z^{a−b} as z → +∞ directly from the integral representation (3), using a Watson-lemma (endpoint) analysis rather than quoting it. The physical payoff is large: this single coefficient Γ(b)/Γ(a) is what makes an unterminated M unnormalizable in a bound-state problem, so its nonvanishing is the mathematical reason quantization exists at all.

Step 1. In the integral (3), for large positive z the factor e^{zt} is sharply largest at the upper endpoint t = 1, so the integral is dominated by a thin boundary layer there:

M(a,b;z)=\frac{\Gamma(b)}{\Gamma(a)\Gamma(b-a)}\int_0^1 e^{zt}\,t^{a-1}(1-t)^{b-a-1}\,\mathrm{d}t,\qquad e^{zt}\ \text{peaks at}\ t=1

Why: unlike a Laplace integral with an interior maximum, here the maximum sits on the boundary, so the leading contribution is an endpoint (Watson-type) analysis, not a saddle point.

Step 2. Magnify the endpoint with t = 1 − s/z, so dt = −ds/z and the s-integration runs over [0, ∞) as z → ∞:

e^{zt}=e^{z}e^{-s},\qquad (1-t)^{b-a-1}=\Big(\frac{s}{z}\Big)^{b-a-1},\qquad t^{a-1}\to 1

Why: rescaling the boundary layer to width 1/z turns the slowly-varying factors into constants to leading order and exposes a clean gamma integral in s.

Step 3. Collect the pieces. The prefactor times e^z z^{−(b−a)} multiplies the standard integral, which is Γ(b − a):

\frac{\Gamma(b)}{\Gamma(a)\Gamma(b-a)}\,e^{z}z^{-(b-a)}\int_0^\infty e^{-s}s^{b-a-1}\,\mathrm{d}s=\frac{\Gamma(b)}{\Gamma(a)\Gamma(b-a)}\,e^{z}z^{a-b}\,\Gamma(b-a)

Why: the ∫ e^{−s} s^{b−a−1} ds = Γ(b − a) cancels the Γ(b − a) in the prefactor exactly, which is the small miracle that leaves a clean coefficient.

Step 4. Read off the result:

M(a,b;z)\sim\frac{\Gamma(b)}{\Gamma(a)}\,e^{z}z^{a-b}\qquad (z\to+\infty)

Why: this is the dominant term of (6); the subdominant Γ(b)/Γ(b − a) (−z)^{−a} piece comes from the lower endpoint t = 0 and only shows itself off the positive real axis, where the Stokes phenomenon controls its weight. *On the elementary case a = 1, b = 2 the leading term gives e^z/z, matching M(1,2;z) = (e^z − 1)/z to O(1/z).*

Go deeper · Derivation C · 2 minHydrogen's radial equation is Kummer's equation

Derivation CHydrogen's radial equation is Kummer's equation

The goal is to show that the single most important eigenvalue problem in physics — the bound states of the hydrogen atom — is Kummer's equation in disguise, and that the Bohr spectrum E_n = −1/(2n²) is nothing but the statement that the confluent series must terminate. We take the radial Schrödinger equation in atomic units, peel off the correct behaviour at both endpoints to land on equation (1), and demand a normalizable solution — which forces a = −n_r and quantizes the energy.

Step 1. With u(r) = r R(r) the Coulomb radial equation, in Hartree atomic units (ħ = m = e = 4πε₀ = 1) with E < 0 for bound states, is

u''+\Big[2E+\frac{2}{r}-\frac{\ell(\ell+1)}{r^2}\Big]u=0

Why: this is the raw problem; the two singular endpoints r → 0 and r → ∞ are where the physics of regularity and normalizability will be imposed.

Step 2. Introduce the decay scale and a dimensionless variable, E = −½κ² and ρ = 2κr:

E=-\tfrac12\kappa^2,\qquad \rho=2\kappa r

Why: writing E in terms of κ makes the asymptotic decay e^{−κr} explicit before we solve anything, so the residual equation only has to describe the polynomial part.

Step 3. Factor out the endpoint behaviour, u = ρ^{ℓ+1} e^{−ρ/2} f(ρ):

u=\rho^{\ell+1}e^{-\rho/2}f(\rho)

Why: ρ^{ℓ+1} enforces regularity at the origin (the centrifugal barrier) and e^{−ρ/2} enforces decay at infinity, so whatever is left for f is the finite, well-behaved core of the wavefunction.

Step 4. Substituting gives an equation for f in exactly Kummer's form,

\rho f''+(2\ell+2-\rho)f'-\Big(\ell+1-\tfrac{1}{\kappa}\Big)f=0

Why: comparing with (1) we read off b = 2ℓ + 2 and a = ℓ + 1 − 1/κ, so f(ρ) = M(a, b; ρ) — the radial function is a confluent hypergeometric function outright.

Step 5. Impose normalizability. If a is not a non-positive integer, M grows like e^ρ and destroys the e^{−ρ/2} decay, so the series must terminate:

a=\ell+1-\frac{1}{\kappa}=-n_r,\quad n_r=0,1,2,\dots\;\Rightarrow\;\frac{1}{\kappa}=n_r+\ell+1\equiv n

Why: this is the whole of quantization — the discrete spectrum is the arithmetic price of cutting off an otherwise-exploding series, and n = n_r + ℓ + 1 is the principal quantum number.

Step 6. Read off the energies:

E_n=-\tfrac12\kappa^2=-\frac{1}{2n^2}\ \text{Ha}

Why: the exact hydrogen spectrum drops out purely from termination, and the radial functions are the terminating M(−n_r, 2ℓ + 2; 2r/n), i.e. associated Laguerre polynomials L^{(2ℓ+1)}_{n−ℓ−1}. *The result reproduces the hydrogenic spectrum $E_n = -1/(2n^2)$ Ha and the standard radial functions $R_{n\ell} \propto r^\ell e^{-r/n} M(\ell+1-n,\, 2\ell+2;\, 2r/n)$ .*

Limiting cases

Nearly every named special function of physics is a confluent hypergeometric function with special parameters — the reason this article is a hub rather than a leaf.

Case	Result	Meaning
Bessel	$J_\nu(z)=\dfrac{(z/2)^\nu}{\Gamma(\nu+1)}\,e^{-iz}\,M\!\left(\nu+\tfrac12,\,2\nu+1;\,2iz\right)$	Bessel of the first kind is a confluent function with imaginary argument; the prefactor strips the regular-singular power $z^\nu$ .
Laguerre	$L_n^{(\alpha)}(z)=\dbinom{n+\alpha}{n}\,M(-n,\,\alpha+1;\,z)$	Terminating $a=-n$ gives the associated Laguerre polynomials — the Coulomb radial bound states.
Hermite	$H_{2n}(z)=(-1)^n\dfrac{(2n)!}{n!}\,M\!\left(-n,\,\tfrac12;\,z^2\right)$	Harmonic-oscillator eigenfunctions; the odd-index ones use $z\,M(-n,\tfrac32;z^2)$ .
Error / incomplete gamma	$\operatorname{erf}(z)=\dfrac{2z}{\sqrt{\pi}}\,M\!\left(\tfrac12,\tfrac32;-z^2\right),\quad \gamma(a,z)=a^{-1}z^a\,M(a,a+1;-z)$	The error function and lower incomplete gamma are single-parameter Kummer functions.
Coulomb wave	$F_\ell(\eta,\rho)=C_\ell(\eta)\,\rho^{\ell+1}e^{-i\rho}\,M\!\left(\ell+1-i\eta,\,2\ell+2;\,2i\rho\right)$	The regular Coulomb function; $C_\ell(\eta)$ is the Gamow normalization built from $\Gamma$ functions.

Common trapA confluent series that does not terminate is still a perfectly good bound-state candidate

It is not. This is the trap that quantization exists to close. If you solve the hydrogen radial equation and keep the full, non-terminating $M(a,b;\rho)$ , the series is finite and convergent for every $\rho$ — nothing looks wrong up close. But by derivation B its large- $\rho$ behaviour is $M\sim \frac{\Gamma(b)}{\Gamma(a)}e^{\rho}\rho^{a-b}$ , and that exponential overwhelms the $e^{-\rho/2}$ you factored out, so $u=\rho^{\ell+1}e^{-\rho/2}M$ actually grows like $e^{+\rho/2}$ and is not normalizable. Only when $a=-n_r$ does the coefficient $\Gamma(b)/\Gamma(a)$ hit a pole of $1/\Gamma(a)$ , killing the exponential and truncating $M$ to a polynomial. Two further cautions: $M(a,b;z)$ is undefined when $b$ is a non-positive integer unless $a$ terminates it first (use $U$ or the regularized $M/\Gamma(b)$ ), and $U(a,b;z)$ carries a branch cut on the negative real axis — never apply real-axis asymptotics blindly off the positive axis.

Worked examples

1 · The hydrogen 2s state, from Kummer's equation to a measured energy

Use the confluent-hypergeometric machinery to build the hydrogen 2s radial function (n = 2, ℓ = 0), locate its radial node, and give its binding energy in electron-volts.

Step 1. Fix the parameters. From derivation C, b = 2ℓ + 2 = 2 and a = ℓ + 1 − n = 0 + 1 − 2 = −1, so a = −n_r with n_r = 1 — the series terminates after two terms. The radial variable is ρ = 2r/n = r (in Bohr radii). So the core function is the polynomial M(−1, 2; r).

Step 2. Evaluate the terminating series. With (−1)₁ = −1 and (2)₁ = 2,

M(-1,2;r)=1+\frac{(-1)_1}{(2)_1}\frac{r}{1!}=1-\frac{r}{2}.

Hence R₂₀(r) ∝ (1 − r/2) e^{−r/2}, the standard 2s radial function.

Step 3. Find the node. The polynomial vanishes when 1 − r/2 = 0, i.e. at r = 2 Bohr radii (r = 2a₀ ≈ 1.06 Å) — the single radial node expected for an n − ℓ − 1 = 1 node state.

Step 4. Get the energy. From E_n = −1/(2n²) Ha with n = 2,

E_2=-\frac{1}{2\cdot 2^2}=-\frac{1}{8}\ \text{Ha}.

Converting with 1 Ha = 27.2114 eV gives E₂ = −27.2114/8 = −3.4014 eV, the observed n = 2 binding energy. The whole state — shape, node, and energy — came out of one terminating Kummer function.

2 · Growth versus decay on an exactly solvable pair

Take a = 1, b = 2, where both confluent functions are elementary. Verify the growth of M and the decay of U at z = 4, and check the leading asymptotic coefficient of derivation B.

Step 1. Identify the functions. From the series, M(1,2;z) = Σ z^n/(n+1)! = (e^z − 1)/z, and from the integral (4), U(1,2;z) = ∫₀^∞ e^{−zt} dt = 1/z exactly (here b − a − 1 = 0 and a − 1 = 0, so the integrand is just e^{−zt}).

Step 2. Evaluate at z = 4. With e⁴ = 54.598,

M(1,2;4)=\frac{e^4-1}{4}=\frac{53.598}{4}=\mathbf{13.40},\qquad U(1,2;4)=\frac14=\mathbf{0.250}.

Step 3. Check the asymptotic law. Derivation B predicts M ~ Γ(b)/Γ(a) · e^z z^{a−b} = Γ(2)/Γ(1) · e^z z^{−1} = e^z/z. At z = 4 that is e⁴/4 = 13.65, within 2% of the exact 13.40, the difference being the −1/z buried in (e^z − 1)/z. Meanwhile U ~ z^{−a} = z^{−1} is exact for this case. M blows up like e^z/z, U decays like 1/z — the two competing behaviours at the irregular point, made completely explicit.

Physics of the portrait

The confluent hypergeometric functions bear two nineteenth-century names — Kummer for the entire solution M, Tricomi for the recessive solution U — and they connect a pure number-theorist's study of a series to the deepest computational tools of quantum mechanics.

Origins · 1 minErnst Kummer and how this was discovered

Ernst Kummer

1810–1893 · the confluent hypergeometric series and Kummer's equation

Ernst Eduard Kummer was one of the towering figures of nineteenth-century German mathematics, and the confluent function that carries his name is only a small corner of his output. Trained in the Gaussian tradition, he taught at a gymnasium in Liegnitz — where Leopold Kronecker was his pupil — before professorships at Breslau and then Berlin, where his students included Georg Cantor, Hermann Schwarz and Georg Frobenius. In an 1836 paper in Crelle's journal he studied the hypergeometric series systematically, and in particular the ₁F₁ series and the transformations (such as M(a,b;z) = e^z M(b−a,b;−z)) that still bear his name. He is equally remembered for inventing "ideal numbers" to rescue unique factorization — the seed of modern ideal theory — for the Kummer surface in algebraic geometry, and for Kummer's convergence test. That a man chasing prime factorizations also handed physicists the exact form of every hydrogen wavefunction is a recurring lesson: the special functions of physics were mapped out, decades early, by pure analysts asking their own questions.

History

1812

— Gauss presents the hypergeometric series ₂F₁ in his Disquisitiones generales

1836

— Kummer publishes his study of the confluent series ₁F₁ and its transformations

1903

— Whittaker recasts the equation in its symmetric Whittaker normal form

1926

— Schrödinger's hydrogen solution turns out to be a terminating ₁F₁

1954

— Tricomi's monograph systematizes the second solution U

Going further

Texts. Arfken, Weber and Harris, Mathematical Methods for Physicists, 7th ed., ch. 13 (hypergeometric and confluent hypergeometric functions) · Stone and Goldbart, Mathematics for Physics, ch. on second-order ODEs and special functions · Riley, Hobson and Bence, Mathematical Methods for Physics and Engineering, special-functions chapter · Abramowitz and Stegun, Handbook of Mathematical Functions, ch. 13 · NIST Digital Library of Mathematical Functions, ch. 13. Key anchor results: the exact pair $M(1,2;z) = (e^z-1)/z$ and $U(1,2;z) = 1/z$ ; the leading asymptotic coefficient $\Gamma(b)/\Gamma(a)$ ; the hydrogen spectrum $E_n = -1/(2n^2)$ Ha, the 2s node at $r = 2a_0$ , and $E_2 = -3.40$ eV. Note that the Kummer transformation is $M(a,b;z) = e^z M(b-a,b;-z)$ — the exponential multiplies the transformed function.

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

Ask the tutor: ask me to specialize M and U to any named function — Bessel, Coulomb, Laguerre, error function — and I will walk you from Kummer's equation to its asymptotics and physics.

What is the difference between M and U, in one sentence each?

M(a, b; z) is Kummer's function: the solution of Kummer's equation that is regular (finite and analytic) at the origin z = 0, given by an everywhere-convergent power series, and which generically grows like e^z at infinity. U(a, b; z) is Tricomi's function: the solution singled out by algebraic decay z^{−a} at +∞, given by a Laplace-type integral, and which generically has a branch-point singularity at the origin. They are the two natural bases for the same two-dimensional solution space, one adapted to the regular singular point and one to the irregular singular point, and the connection formula (5) expresses U as a specific Gamma-weighted combination of M and its second Frobenius partner.

Why is it called "confluent" and confluent with what?

"Confluent" means "flowing together", and it describes what happens to singular points, not to functions. The Gauss hypergeometric equation for ₂F₁ has three regular singular points, at z = 0, 1 and ∞. If you rescale the variable as x = z/β and let β → ∞, the singular point at x = 1 is carried out to z = β and eventually collides with the point already at infinity. Two regular singular points thereby merge — confluunt — into a single irregular singular point of rank one, and the equation that survives the limit is Kummer's. So "confluent hypergeometric" literally means "the hypergeometric function you get after two of its singular points have flowed together".

How does the exponential e^z get into an equation with only polynomial coefficients?

It is the fingerprint of the irregular singular point. At a regular singular point the Frobenius method gives power-law solutions z^ρ fixed by an indicial equation; an irregular singular point admits no such clean local power series, and its solutions carry an essential singularity — here an exponential factor e^z. Because the irregular point at infinity was manufactured by fusing two regular points (the confluent limit), the exponential is the imprint of that fusion. Derivation B shows the e^z emerging from the endpoint of the Kummer integral: for large positive z the integrand piles up at t = 1, and e^{z·1} = e^z is the dominant factor that survives.

Why does demanding normalizability quantize the energy?

Because the generic confluent solution grows too fast to normalize, and only special parameter values tame it. In the hydrogen problem you factor the wavefunction as u = ρ^{ℓ+1} e^{−ρ/2} M(a, b; ρ). The prefactor e^{−ρ/2} handles the decay you want, but M itself generically grows like e^{+ρ} at large ρ (with coefficient Γ(b)/Γ(a)), so the product actually diverges as e^{+ρ/2}. The only escape is to make M a polynomial, which happens exactly when a = −n_r is a non-positive integer — then 1/Γ(a) has a zero, the exponential coefficient vanishes, and the series truncates. That single arithmetic condition, a = −n_r, is the quantization condition, and unwinding it gives n = n_r + ℓ + 1 and E_n = −1/(2n²).

Are Bessel functions really confluent hypergeometric functions?

Yes, exactly, though the parameters are imaginary. The identity J_ν(z) = (z/2)^ν/Γ(ν+1) · e^{−iz} M(ν + ½, 2ν + 1; 2iz) writes the Bessel function of the first kind as a Kummer function with argument 2iz. This is more than bookkeeping: the large-argument asymptotics of J_ν, including the oscillatory √(2/πz) cos(z − …) form and the Stokes phenomenon connecting the Hankel functions H⁽¹⁾ and H⁽²⁾, are just the M/U asymptotics of (6) at imaginary argument. The modified Bessel functions I_ν and K_ν are likewise confluent functions, with K_ν a U-type (recessive) solution — which is why K_ν is the one that decays.

When does the connection formula for U break down, and what fixes it?

The formula (5) has Γ(1 − b) and Γ(b − 1) in it, and both blow up when b is an integer — precisely the case that occurs constantly in physics, since b = 2ℓ + 2 for hydrogen is always an even integer. When b is an integer the two Frobenius solutions at z = 0 are no longer independent (their indicial exponents 0 and 1 − b differ by an integer), and one develops a logarithm. The correct U is then obtained by a careful limit b → integer in which the divergent Gamma factors cancel against the coalescing solutions, producing a log z term — the same integer-resonance subtlety that gives Bessel Y_n its logarithm.