Ordinary and Singular Points

A linear second-order ODE with variable coefficients almost never surrenders a closed-form solution, yet its local behaviour is completely legible from a single glance at the coefficient functions. Write the homogeneous equation in standard form $y''+p(x)y'+q(x)y=0$ . The one question that organises everything downstream is: are $p$ and $q$ analytic — do they possess a convergent Taylor series — at the point $x_0$ you care about? If yes, $x_0$ is an ordinary point and both solutions are themselves analytic there, expandable as ordinary power series whose radius of convergence you can predict before writing a single coefficient. If no, $x_0$ is singular, and the sharper question becomes how violently the coefficients diverge. That refinement is the heart of the subject and the achievement of Lazarus Fuchs: a regular singular point tolerates at worst a simple pole in $p$ and a double pole in $q$ — equivalently $(x-x_0)p$ and $(x-x_0)^2q$ stay analytic — and this precise, borderline rate of blow-up is exactly what a generalised power series, the Method of Frobenius, is built to absorb. Anything worse is an irregular singular point, where convergent series die and one must reach for asymptotics or WKB. Nearly every named equation of physics — Bessel, Legendre, Laguerre, Hermite, hypergeometric — lives entirely on ordinary and regular singular points in the finite plane, which is precisely why their solutions became the "special functions" tabulated for two centuries. The point at infinity joins the same scheme through the substitution $t=1/x$ . Master this taxonomy and you know, in advance, which of three tools an unfamiliar equation demands.

Key takeaways

Put the ODE in standard form $y''+p(x)y'+q(x)y=0$ first; the whole classification reads off the analyticity of $p$ and $q$ — and the singular points hide at the zeros of the leading coefficient you divided by.
$x_0$ is an ordinary point if $p$ and $q$ are both analytic there; then two independent power-series solutions exist with radius of convergence at least the distance to the nearest singularity in the complex plane.
$x_0$ is a regular singular point if $p$ or $q$ fails, but $(x-x_0)p$ and $(x-x_0)^2q$ are both analytic — the mild case where Frobenius series $y=(x-x_0)^s\sum a_n(x-x_0)^n$ succeed.
If even those products blow up, $x_0$ is an irregular singular point: power series and Frobenius generically fail, and solutions carry essential singularities like $e^{1/x}$ .
The class picks the method: ordinary power series, Frobenius, or asymptotic analysis — and at a regular singular point the exponents $s$ come from the indicial equation $s(s-1)+p_0s+q_0=0$ .
The point at infinity is classified by setting $t=1/x$ and examining the transformed equation at $t=0$ .
Key idea: In $y''+p(x)y'+q(x)y=0$ classify $x_0$ by analyticity — ordinary if $p,q$ are analytic, regular singular if $(x-x_0)p$ and $(x-x_0)^2q$ are analytic, irregular singular otherwise — and the class fixes whether ordinary power series, Frobenius series, or asymptotic methods apply.

The core insight

In $y''+p(x)y'+q(x)y=0$ classify $x_0$ by analyticity: ordinary if $p,q$ are analytic; regular singular if $(x-x_0)p$ and $(x-x_0)^2q$ are analytic; irregular singular otherwise. The class determines whether ordinary power series, Frobenius series, or asymptotic methods apply.

Where this lives today

The taxonomy is not a museum piece; it is the front-end of every computer algebra system's ODE solver. When Maple, Mathematica, or SymPy is handed a second-order linear equation, the first thing it does is locate the singular points and classify each as ordinary, regular singular, or irregular — that verdict routes the problem to power-series, Frobenius, or asymptotic (formal/Newton-polygon) machinery, and decides whether a closed-form "hypergeometric-type" answer is even possible. In physics the same reflex governs quantum mechanics: solving the Schrödinger equation for hydrogen or the harmonic oscillator means recognising that infinity is an irregular singular point where one must discard the exponentially growing solution, and demanding the Frobenius series at the origin terminate — the two boundary demands together quantise the energy. Modern research on the exact WKB method, resurgence, and the Stokes phenomenon is essentially a deep study of what happens at irregular singular points, where the divergent asymptotic series near infinity encode, through their Borel sums, the tunnelling and non-perturbative effects that ordinary perturbation theory misses. Confluent limits — merging two regular singular points into one irregular point — connect the hypergeometric function to the Coulomb, Bessel, and parabolic-cylinder functions that fill the tables of atomic and nuclear physics.

The physical picture

Think of $p(x)$ and $q(x)$ as the two forces steering a solution curve. Where they vary smoothly the steering is gentle and the solution inherits that smoothness — it has a Taylor series and can be built coefficient by coefficient. Where a coefficient diverges the steering becomes infinitely strong over a vanishing interval, and the solution may spiral, branch, or develop its own singularity. The subtlety that makes the theory beautiful is that "nearest singularity" must be measured in the complex plane, not just along the real axis: a coefficient perfectly smooth for every real $x$ can still cap the reach of a real power series, because its complex singularities are what the series feels first.

Figure

Radius of convergence is set by the nearest singularity in the complex plane. For (1+x²)y'' + y = 0 the coefficient q = 1/(1+x²) is smooth for every real x, yet a power series about the ordinary point x = 0 converges only for x less than 1, because complex singularities sit at x = ±i, a distance R = 1 away.

The regular/irregular distinction measures how fast a coefficient blows up. A regular singular point permits $p\sim 1/x$ and $q\sim 1/x^2$ — the exact borderline at which, after multiplying the equation by $(x-x_0)^2$ , the coefficients become bounded and the equation reduces in leading order to an Euler equation $x^2y''+bxy'+cy=0$ whose solutions are pure powers $x^s$ . That Euler skeleton is the solvable model a Frobenius series perturbs, and the exponents $s$ it demands are the roots of the indicial equation. An irregular singular point is where the natural local model is no longer $x^s$ but something like $e^{f(x)}$ with $f$ itself singular — an essential singularity that no convergent power series can reproduce. The decision procedure is a short flowchart, and the second figure is that flowchart with Bessel's equation walked through it.

Figure

The classification decision tree for y'' + p y' + q y = 0 at a point x₀, with Bessel's equation x²y'' + xy' + (x² − ν²)y = 0 traced through it: at x = 0 the products xp = 1 and x²q = x² − ν² are analytic so the origin is regular singular, while at x = ∞ the transform t = 1/x gives t²q ~ 1/t², not analytic, so infinity is irregular singular.

The core equations

Everything begins with a general homogeneous linear second-order equation. Its coefficients $a_2,a_1,a_0$ are usually polynomials, and the whole classification depends on where the leading coefficient $a_2$ vanishes — those zeros are the only places (short of genuine singularities of $a_1,a_0$ ) where trouble can arise.

a_2(x)\,y'' + a_1(x)\,y' + a_0(x)\,y = 0 \qquad (1)

where: $a_2,a_1,a_0$ — coefficient functions, assumed analytic (often polynomials); $y(x)$ — the unknown function of the independent variable $x$ . Reading: as written, the classification is invisible; it only appears once you normalise. Dividing through by $a_2$ puts the equation in standard form, exposing the two functions whose analyticity decides everything. This step is not cosmetic — it is where the singular points are born.

y'' + p(x)\,y' + q(x)\,y = 0,\qquad p=\frac{a_1}{a_2},\quad q=\frac{a_0}{a_2} \qquad (2)

where: $p(x)$ — coefficient of $y'$ , tested for analyticity; $q(x)$ — coefficient of $y$ , tested together with $p$ . Reading: wherever $a_2\ne0$ and $a_1,a_0$ are analytic, $p$ and $q$ are ratios of analytic functions and hence analytic; the candidate singular points are precisely the zeros of $a_2$ . The first verdict is the simplest. If both standard-form coefficients are analytic at $x_0$ , the point is ordinary and the Existence–Uniqueness theorem guarantees two analytic solutions.

x_0\ \text{ordinary}\iff p(x)\ \text{and}\ q(x)\ \text{both analytic at }x_0 \qquad (3)

where: $x_0$ — the point being classified. Reading: "analytic" means a convergent Taylor expansion exists in some disc about $x_0$ ; a mere finite value is not enough, but for rational $p,q$ it amounts to the denominator not vanishing. When analyticity fails, Fuchs's criterion sorts the mild singularities from the wild. Multiplying $p$ by one factor of $(x-x_0)$ and $q$ by two factors must repair the analyticity for the point to count as regular.

x_0\ \text{regular singular}\iff (x-x_0)\,p(x)\ \text{and}\ (x-x_0)^2\,q(x)\ \text{both analytic at }x_0 \qquad (4)

where: $(x-x_0)p$ — clears at most a simple pole of $p$ ; $(x-x_0)^2q$ — clears at most a double pole of $q$ . Reading: the asymmetric powers $1$ and $2$ are dictated by the term balance $y''\sim p\,y'\sim q\,y$ ; if either product still diverges, the point is irregular singular. At an ordinary point the solution is an ordinary Taylor series, and its radius of convergence is bounded below by geometry alone.

y(x)=\sum_{n=0}^{\infty} a_n\,(x-x_0)^n,\qquad |x-x_0| < R \qquad (5)

where: $a_n$ — coefficients fixed by a two-term (or longer) recurrence read off the ODE; $R$ — radius of convergence, at least the distance from $x_0$ to the nearest singularity of $p$ or $q$ . Reading: $a_0$ and $a_1$ stay free (they encode $y(x_0)$ and $y'(x_0)$ ), generating the two independent solutions. At a regular singular point the ordinary series is replaced by a Frobenius series — a power series multiplied by a possibly non-integer power $(x-x_0)^s$ that absorbs the mild singularity.

y(x)=(x-x_0)^{s}\sum_{n=0}^{\infty} a_n\,(x-x_0)^n,\qquad a_0\ne 0 \qquad (6)

where: $s$ — the indicial exponent, a root of the indicial equation, generally not an integer; $a_0$ — leading coefficient, normalised nonzero so that $s$ genuinely captures the leading power. Reading: with $s=0$ this collapses to an ordinary series, so (6) contains (5) as a special case. The exponent $s$ is fixed before any recurrence, by demanding the lowest-order terms balance against the Euler skeleton. That balance is the indicial equation.

s(s-1) + p_0\,s + q_0 = 0 \qquad (7)

where: $p_0=\lim_{x\to x_0}(x-x_0)p(x)$ — the leading Laurent coefficient of $p$ ; $q_0=\lim_{x\to x_0}(x-x_0)^2q(x)$ — the leading coefficient of $q$ . Reading: this quadratic has roots $s_1,s_2$ ; when they coincide or differ by a positive integer, a logarithm generically enters the second solution. Finally, the point at infinity is folded into the same machinery by the inversion $t=1/x$ , which maps large $x$ to small $t$ and lets the finite-point criteria be applied at $t=0$ .

t=\frac{1}{x},\qquad \frac{\mathrm{d}}{\mathrm{d}x}=-t^2\frac{\mathrm{d}}{\mathrm{d}t},\qquad \frac{\mathrm{d}^2}{\mathrm{d}x^2}=t^4\frac{\mathrm{d}^2}{\mathrm{d}t^2}+2t^3\frac{\mathrm{d}}{\mathrm{d}t} \qquad (8)

where: $t$ — the new variable, with $x\to\infty$ corresponding to $t\to0$ . Reading: substitute (8) into the ODE, collect into standard form in $t$ , and classify $t=0$ by criteria (3)–(4).

Derivation AThe recurrence and the guaranteed radius of convergence at an ordinary point

The goal is to show that expanding a solution in an ordinary power series about an ordinary point is never a leap of faith: the substitution converts the ODE into an algebraic recurrence for the coefficients, exactly two of which stay free, and the analyticity of $p$ and $q$ guarantees the resulting series converges out to at least the distance to the nearest singularity. We do this concretely with $x_0=0$ ordinary, so that $p=\sum_k p_k x^k$ and $q=\sum_k q_k x^k$ converge in a disc $|x|<R$ , and we finish with a numerical check that reproduces $\cos x$ and $\sin x$ .

Step 1. Insert $y=\sum_{n\ge0}a_nx^n$ , differentiate term by term to get $y'=\sum_{n\ge1}na_nx^{n-1}$ and $y''=\sum_{n\ge0}(n+2)(n+1)a_{n+2}x^n$ , and multiply the series for $p,q$ into those for $y',y$ as Cauchy products. Why: term-by-term differentiation is legal strictly inside the radius of convergence, which is exactly the region an ordinary point guarantees us.

Step 2. Collect the coefficient of $x^n$ and set it to zero, giving $(n+2)(n+1)a_{n+2}+\sum_k\big[p_k(n-k+1)a_{n-k+1}+q_k a_{n-k}\big]=0$ , which solves for $a_{n+2}$ in terms of all lower coefficients. Why: a power series vanishes identically only if every coefficient vanishes, so one functional equation becomes infinitely many algebraic ones — the recurrence.

Step 3. Observe that $a_0$ and $a_1$ never get pinned by the recurrence; they remain arbitrary and set $y(0)$ and $y'(0)$ . Choosing $(a_0,a_1)=(1,0)$ and $(0,1)$ builds two solutions whose Wronskian at $x=0$ is $1\ne0$ , so they are linearly independent. Why: a second-order equation carries exactly two degrees of freedom, and the two free coefficients are precisely those constants of integration.

Step 4. Bound the growth: because $p,q$ are analytic in $|x|<R$ their coefficients satisfy $|p_k|,|q_k|\le M/\rho^k$ for any $\rho<R$ , and a standard majorant (dominating-series) argument then forces $|a_n|$ to grow no faster than $\rho^{-n}$ , so the solution series converges for $|x|<R$ as well. Why: this is Fuchs's theorem — the solution can only misbehave where a coefficient does, so the nearest singularity of $p,q$ caps, but never undercuts, the radius. *Verified: for $y''+y=0$ we have $p=0,q=1$ entire, so $R=\infty$ ; the recurrence collapses to $a_{n+2}=-a_n/[(n+2)(n+1)]$ , giving $a_0(1-\tfrac{x^2}{2!}+\tfrac{x^4}{4!}-\cdots)+a_1(x-\tfrac{x^3}{3!}+\cdots)=a_0\cos x+a_1\sin x$ , both with infinite radius.*

Go deeper · Derivation B · 2 minThe indicial equation from the Euler skeleton

Derivation BThe indicial equation from the Euler skeleton

The aim here is to explain why a regular singular point comes with a quadratic — the indicial equation — that fixes the leading exponent $s$ before any coefficient is computed, and why the specific combination $s(s-1)+p_0s+q_0$ appears. The engine is that near a regular singular point the equation, stripped to leading order, is an Euler (equidimensional) equation whose solutions are exact powers $x^s$ ; the full equation is an analytic perturbation of that solvable model, so its Frobenius series must begin with the same power. We take $x_0=0$ without loss of generality.

Step 1. Multiply the standard form by $x^2$ to get $x^2y''+x\,(xp)\,y'+(x^2q)\,y=0$ , and write $xp=p_0+p_1x+\cdots$ and $x^2q=q_0+q_1x+\cdots$ , which are analytic precisely because the point is regular singular. Why: the factors $x$ and $x^2$ are chosen so that each of the three terms carries the same net power of $x$ , the defining feature of an Euler equation, so they can balance at leading order.

Step 2. Keep only the constant parts $p_0,q_0$ ; the leading model is the Euler equation $x^2y''+p_0xy'+q_0y=0$ . Try $y=x^s$ : then $x^2y''=s(s-1)x^s$ , $p_0xy'=p_0sx^s$ , $q_0y=q_0x^s$ . Why: an equidimensional equation is scale-invariant, and the only functions reproduced up to a constant by $x\,\mathrm{d}/\mathrm{d}x$ are pure powers, so $x^s$ is the natural trial solution.

Step 3. Sum the three contributions: $[\,s(s-1)+p_0s+q_0\,]x^s=0$ . Since $x^s\ne0$ , the bracket must vanish, which is the indicial equation $(7)$ ; its roots $s_1,s_2$ are the two admissible leading exponents. Why: this is the $n=0$ term of the full Frobenius substitution — with $a_0\ne0$ it is the lowest power present, so its coefficient must vanish on its own, decoupled from every higher term.

Step 4. Read the outcome. If $s_1-s_2$ is not an integer, two independent Frobenius series exist. If $s_1=s_2$ , only one does and the second solution acquires a $\log$ term. If $s_1-s_2$ is a positive integer, the smaller root may or may not need a logarithm, decided by whether a resonance in the recurrence is hit. Why: the recurrence for exponent $s$ has denominators $\propto(s+n)(s+n-1)+p_0(s+n)+q_0$ , which vanish exactly when $s+n$ equals the other root — an integer gap is what triggers the breakdown that a logarithm repairs. *Verified: for Bessel with $p_0=1,q_0=-\nu^2$ the indicial equation is $s^2-\nu^2=0$ , giving $s=\pm\nu$ — the exponents of $J_\nu$ and $J_{-\nu}$ ; for a double root $\nu=0$ the second solution $Y_0$ indeed carries a $\log x$ .*

Go deeper · Derivation C · 2 minClassifying Bessel's equation at the origin and at infinity

Derivation CClassifying Bessel's equation at the origin and at infinity

Bessel's equation is the canonical specimen carrying one regular singular point (the origin) and one irregular singular point (infinity), and working it end to end exercises every tool: standard form, the Fuchs products, and the $t=1/x$ inversion for the point at infinity. The payoff is a complete prediction — Frobenius series converge at the origin with exponents $\pm\nu$ , while at infinity no convergent series exists and the solutions must be described asymptotically, matching the known $x^{-1/2}\cos(x-\phi)$ tail of $J_\nu$ .

Step 1. Divide Bessel's equation $x^2y''+xy'+(x^2-\nu^2)y=0$ by $x^2$ to reach standard form $y''+\tfrac1x y'+\tfrac{x^2-\nu^2}{x^2}y=0$ , so $p=1/x$ has a simple pole and $q=1-\nu^2/x^2$ has a double pole at the origin. Why: only in standard form do the poles of $p,q$ — the objects the criteria test — become visible; the leading coefficient $x^2$ vanishing at $0$ is the tell that the origin is singular.

Step 2. Form the Fuchs products: $x\,p=1$ and $x^2q=x^2-\nu^2$ , both polynomials and hence analytic at $x=0$ . Therefore the origin is a regular singular point, with $p_0=1$ and $q_0=-\nu^2$ , giving indicial roots $s=\pm\nu$ . Why: the single factor of $x$ exactly cancels the simple pole and the two factors cancel the double pole; that they leave polynomials, not new poles, is the definition of "regular".

Step 3. Classify infinity by substituting $t=1/x$ using $(8)$ . Then $x^2y''=t^2\ddot y+2t\dot y$ , $xy'=-t\dot y$ , and $(x^2-\nu^2)y=(t^{-2}-\nu^2)y$ ; summing and dividing by $t^2$ gives $\ddot y+\tfrac1t\dot y+\big(\tfrac1{t^4}-\tfrac{\nu^2}{t^2}\big)y=0$ . Why: infinity is not a point you can test directly, but inversion turns "how do coefficients behave as $x\to\infty$ " into the familiar "how do they behave as $t\to0$ ".

Step 4. Apply the criteria at $t=0$ : here $\tilde p=1/t$ so $t\tilde p=1$ is fine, but $\tilde q\sim1/t^4$ so $t^2\tilde q\sim1/t^2$ still diverges. The second Fuchs product fails, so $t=0$ — that is, $x=\infty$ — is an irregular singular point. Why: a $1/t^4$ coefficient blows up two orders faster than the regular borderline $1/t^2$ , so no finite power $t^s$ can lead the solution; the true behaviour is the essential-singularity oscillation $e^{\pm i x}$ seen in the asymptotics of $J_\nu$ . *Verified: standard references (Arfken–Weber; Watson) confirm $J_\nu$ is an analytic Frobenius solution at $0$ with index $\nu$ , while at infinity $J_\nu(x)\sim\sqrt{2/(\pi x)}\cos(x-\nu\pi/2-\pi/4)$ is an asymptotic, non-convergent expansion — exactly the irregular-point signature.*

Limiting cases

The classification of $x_0=0$ (any point maps here by a shift) is governed entirely by the leading singular behaviour of $p$ and $q$ :

Case	Result	Meaning
Ordinary point	$p(x),\,q(x)$ analytic at $0$	Two analytic solutions; use ordinary power series with $R\ge$ distance to nearest singularity. Example: $y''+y=0$ at any point ( $R=\infty$ ).
Regular singular point	$xp(x),\,x^2q(x)$ analytic at $0$	At worst $p\sim1/x$ , $q\sim1/x^2$ ; the Euler skeleton survives and Frobenius applies with exponents from $s(s-1)+p_0s+q_0=0$ . Example: Bessel at $x=0$ , indices $\pm\nu$ .
Irregular singular point	$xp(x)$ or $x^2q(x)$ not analytic	Coefficients diverge faster than the Euler borderline; Frobenius fails, solutions carry essential singularities, use asymptotics/WKB. Example: $y''+x^{-3}y=0$ at $x=0$ .
Point at infinity	$t=1/x$ , classify the transformed equation at $t=0$	Same three verdicts, applied after inversion. Example: Bessel at $x=\infty$ is irregular.

Common trapthe misconception

The single most common error is testing analyticity before normalising. The criteria refer to $p=a_1/a_2$ and $q=a_0/a_2$ , so you must divide by the leading coefficient $a_2(x)$ first — and its zeros are exactly the candidate singular points that vanish from view if you skip this step. A close second trap is confusing a singular point of the equation with a singular point of a solution: at a regular singular point one Frobenius solution is often perfectly finite there (Legendre's $P_\ell$ is analytic at $x=\pm1$ ), while its partner diverges — the coefficients being singular only means some solution misbehaves, not all of them. A third: "regular singular" still means singular; the word "regular" only promotes it above "irregular", it does not make the point ordinary.

Worked examples

1 · Hermite's equation — an ordinary point, a two-term recurrence, and quantisation (fully worked)

Classify $x=0$ for Hermite's equation $y''-2xy'+2Ny=0$ , find the recurrence for the power-series coefficients, and show what forces the series to terminate.

Step 1. Put it in standard form: it already is, with $p=-2x$ and $q=2N$ . Both are polynomials, hence entire (analytic for all finite $x$ ), so every finite point is ordinary and there is no finite singularity — the guaranteed radius of convergence is $R=\infty$ .

Step 2. Insert $y=\sum_{n\ge0}a_nx^n$ . The three pieces are $y''=\sum_n(n+2)(n+1)a_{n+2}x^n$ , $-2xy'=\sum_n(-2n)a_nx^n$ , and $2Ny=\sum_n 2N a_nx^n$ . Collecting the coefficient of $x^n$ gives $(n+2)(n+1)a_{n+2}-2na_n+2Na_n=0$ .

Step 3. Solve for the recurrence: $a_{n+2}=\dfrac{2(n-N)}{(n+1)(n+2)}\,a_n$ . It links $a_{n+2}$ to $a_n$ only, so the even coefficients (built from $a_0$ ) and odd coefficients (built from $a_1$ ) form two independent solutions — as required at an ordinary point.

Step 4. Read off termination: the numerator $2(n-N)$ vanishes at $n=N$ , so if $N$ is a non-negative integer, $a_{N+2}=0$ and every coefficient beyond it dies — the surviving series is a degree- $N$ polynomial, the Hermite polynomial $H_N$ (up to normalisation). For non-integer $N$ the series never terminates and grows like $e^{x^2}$ , blowing up at infinity. Demanding a solution that does not blow up quantises $N$ to a non-negative integer — this is exactly how the quantum harmonic-oscillator energies $E=\hbar\omega(N+\tfrac12)$ arise.

2 · Legendre's equation — regular singular endpoints and a double indicial root

Classify the points of Legendre's equation $(1-x^2)y''-2xy'+\ell(\ell+1)y=0$ and find the indicial exponents at $x=1$ .

Step 1. Standard form: $y''-\dfrac{2x}{1-x^2}y'+\dfrac{\ell(\ell+1)}{1-x^2}y=0$ . The leading coefficient $1-x^2=(1-x)(1+x)$ vanishes only at $x=\pm1$ , so those are the only candidate singular points; every $x\in(-1,1)$ is ordinary.

Step 2. Test $x=1$ with the Fuchs products. Using $1-x^2=-(x-1)(1+x)$ , (x-1)p=(x-1)\cdot\frac{-2x}{-(x-1)(1+x)}=\frac{2x}{1+x},\qquad (x-1)^2q=(x-1)^2\cdot\frac{\ell(\ell+1)}{-(x-1)(1+x)}=\frac{-(x-1)\ell(\ell+1)}{1+x}. Both are analytic at $x=1$ (denominators $\to2$ ), so $x=1$ is a regular singular point — and by symmetry so is $x=-1$ .

Step 3. Extract the indicial data as the values of those products at $x=1$ : $p_0=\dfrac{2\cdot1}{1+1}=1$ and $q_0=0$ . The indicial equation $(7)$ is therefore $s(s-1)+1\cdot s+0=s^2=0$ .

Step 4. The double root $s=0$ means the two exponents coincide, so only one Frobenius solution is a plain power series — the Legendre polynomial $P_\ell$ , finite at $x=\pm1$ — while the second solution carries a logarithm and is the function $Q_\ell$ , which diverges logarithmically at $x=\pm1$ . The interior expansion about $x=0$ converges for $|x|<1$ , exactly the distance to those endpoints. Result: $x=\pm1$ regular singular, indices $s=0,0$ , second solution logarithmic.

3 · The hypergeometric equation — three regular singular points organise the special functions

Show that the hypergeometric equation $x(1-x)y''+[c-(a+b+1)x]y'-ab\,y=0$ has regular singular points at $x=0$ , $x=1$ , and $x=\infty$ , and no others.

Step 1. Standard form: $p=\dfrac{c-(a+b+1)x}{x(1-x)}$ and $q=\dfrac{-ab}{x(1-x)}$ . The leading coefficient $x(1-x)$ vanishes only at $x=0$ and $x=1$ , the finite candidates.

Step 2. At $x=0$ : $x\,p=\dfrac{c-(a+b+1)x}{1-x}\to c$ and $x^2q=\dfrac{-abx}{1-x}\to0$ , both analytic — regular singular, with indicial roots from $s(s-1)+cs=0$ , i.e. $s=0$ and $s=1-c$ .

Step 3. At $x=1$ : $(x-1)p\to\dfrac{(a+b+1)-c}{1}$ (finite) and $(x-1)^2q\to0$ , both analytic — regular singular, with exponents $0$ and $c-a-b$ .

Step 4. At infinity, set $t=1/x$ ; the transformed coefficients give finite Fuchs products with indicial roots $a$ and $b$ — regular singular too. So the hypergeometric equation has exactly three regular singular points, at $0,1,\infty$ , and any second-order linear ODE with three regular singular points can be mapped onto it — which is why Legendre, Chebyshev, Jacobi, and many others are all special cases of the single hypergeometric function.

Physics of the portrait

The classification of singular points is one of those rare pieces of pure analysis that arrived exactly when physics needed it: by the 1860s the equations of Bessel, Legendre, and Gauss were everywhere in potential theory and wave problems, but there was no general principle saying why their series solutions worked. Lazarus Fuchs supplied it.

Origins · 1 minLazarus Fuchs and how this was discovered

Lazarus Fuchs

1833–1902 · the analytic theory of linear differential equations

Lazarus Immanuel Fuchs, a student of Weierstrass in Berlin, transformed the study of linear ODEs from a catalogue of clever tricks into a systematic theory built on complex analysis. In a series of papers in Crelle's Journal around 1865–1866 he asked the decisive question: given a linear equation whose coefficients are analytic except at isolated points, near which of those points do all solutions remain "regular" — bounded by a finite power of the distance — rather than exploding with an essential singularity? His answer is the criterion that now bears his name: the point $x_0$ admits only regular solutions precisely when $(x-x_0)p$ and $(x-x_0)^2q$ are analytic there. Equations all of whose singular points (including infinity) are of this benign type are called Fuchsian to this day, and the hypergeometric equation is the three-singular-point archetype. Fuchs showed how to read the leading exponents off an indicial equation, clarifying when a second solution must acquire a logarithm — the machinery Frobenius would soon cast into a practical series algorithm. His work fed directly into Riemann's monodromy viewpoint and, through Poincaré and Klein, into the theory of automorphic functions; Poincaré originally called the functions he discovered "Fuchsian" in tribute. What Fuchs gave physics is the confidence to expand: knowing in advance that a coefficient's worst singularity is a simple or double pole tells you a convergent (Frobenius) series exists before you compute a single term.

History

1857

— Riemann characterises the hypergeometric equation entirely by its three regular singular points and their exponents

1866

— Fuchs proves that solutions stay regular exactly when (x−x₀)p and (x−x₀)²q are analytic, defining regular singular points

1873

— Frobenius turns the indicial analysis into a systematic power-series algorithm y = (x−x₀)ˢ Σ aₙ(x−x₀)ⁿ

1886

— Poincaré founds the theory of asymptotic expansions, the right language for irregular singular points

1909

— Birkhoff develops the general theory of irregular singular points and their formal solutions

Going further

Texts. Arfken & Weber, Mathematical Methods for Physicists, "Series Solutions — Frobenius's Method" · Hassani, Mathematical Physics, chapters on ODEs in the complex domain · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, "Series solutions of ODEs" · Stone & Goldbart, Mathematics for Physics, ODE and Sturm–Liouville chapters · Whittaker & Watson, A Course of Modern Analysis, Fuchs theory and the hypergeometric equation · Bender & Orszag, Advanced Mathematical Methods for Scientists and Engineers, local analysis of irregular points.

Provenance. The standard-form reduction and the ordinary/regular/irregular criteria follow Arfken–Weber and Riley–Hobson–Bence; the Fuchs regular/irregular distinction and the indicial equation follow Whittaker & Watson and Stone–Goldbart; the asymptotic treatment of irregular points follows Bender–Orszag. A verifier should recompute: (i) the indicial equation $s(s-1)+p_0s+q_0=0$ from the Euler skeleton; (ii) the Bessel classification — $xp=1$ , $x^2q=x^2-\nu^2$ analytic at $0$ (regular, indices $\pm\nu$ ), and the $t=1/x$ transform giving $t^2\tilde q\sim t^{-2}$ at infinity (irregular); (iii) the Legendre products at $x=1$ giving $p_0=1$ , $q_0=0$ , indices $s=0,0$ ; (iv) the Hermite recurrence $a_{n+2}=2(n-N)/[(n+1)(n+2)]\,a_n$ and its termination at $n=N$ ; (v) the three regular singular points of the hypergeometric equation. All exponents, radii, and recurrences are standard textbook results.

Prerequisites. Linear Second-Order ODEs · Power Series Solutions · Existence and Uniqueness Theorems for ODEs. Leads to. The Method of Frobenius · The Indicial Equation and Exponents · Sturm–Liouville Theory.

Ask the tutor: Hand me any second-order linear ODE and I will normalise it, hunt down every singular point including infinity, classify each as ordinary, regular, or irregular, and tell you whether to reach for a power series, Frobenius, or asymptotics.

If a point is a "regular singular point", is it singular or not?

It is genuinely singular — at least one of $p,q$ fails to be analytic there, and generically at least one solution of the equation misbehaves (diverges or branches) at that point. The adjective "regular" is a promotion within the singular class: it says the singularity is the mild, borderline kind that a Frobenius series can handle, as opposed to an "irregular" singular point where even Frobenius fails. A useful mental ladder is ordinary (both solutions analytic) → regular singular (solutions at worst algebraic-times-log) → irregular singular (essential singularities). So "regular singular" sits in the middle: singular enough that ordinary power series break, tame enough that a generalised power series still works.

Why exactly one factor of (x−x₀) for p but two for q?

Because those are the powers that keep the three terms of the equation in balance. In $y''+py'+qy=0$ , a candidate leading behaviour $y\sim(x-x_0)^s$ makes $y''\sim(x-x_0)^{s-2}$ , $py'\sim p\,(x-x_0)^{s-1}$ , and $qy\sim q\,(x-x_0)^s$ . For all three to scale as $(x-x_0)^{s-2}$ — so that a single indicial equation can balance them — you need $p$ no worse than $(x-x_0)^{-1}$ and $q$ no worse than $(x-x_0)^{-2}$ . Multiplying by $(x-x_0)$ and $(x-x_0)^2$ respectively is just the test that those bounds hold. Any milder singularity than the borderline still passes; anything stronger ruins the balance and defines an irregular point.

How can a solution be analytic at a point where the equation is singular?

Very commonly, and Legendre is the clean example: $x=\pm1$ are regular singular points of the equation, yet the Legendre polynomials $P_\ell$ are perfectly finite and analytic there — it is the second solution $Q_\ell$ that carries the logarithmic singularity. A singular point of the equation only tells you that the pair of solutions cannot both be analytic; the indicial exponents say which behaviours are available. When one indicial root is a non-negative integer, the corresponding Frobenius solution reduces to an ordinary Taylor series and is analytic at the point. This is why physical boundary conditions ("stay finite at the pole", "regular at the origin") often single out one specific solution — the analytic one — and thereby quantise the problem.

What is the practical recipe for classifying the point at infinity?

Substitute $t=1/x$ and $\mathrm{d}/\mathrm{d}x=-t^2\,\mathrm{d}/\mathrm{d}t$ , $\mathrm{d}^2/\mathrm{d}x^2=t^4\,\mathrm{d}^2/\mathrm{d}t^2+2t^3\,\mathrm{d}/\mathrm{d}t$ , into the equation, then rewrite it in standard form in $t$ and classify $t=0$ by the ordinary rules. A quick shortcut: after transforming, infinity is an ordinary point only if the new $\tilde p(t)$ behaves like $2/t+\cdots$ minus analytic terms and $\tilde q(t)$ is analytic — most equations are actually singular at infinity even when they look tame. Bessel is irregular at infinity; Legendre and the hypergeometric equation are regular singular there. Always check infinity explicitly: a global count of singular points (finite plus infinity) is what tells you, for instance, that the hypergeometric equation has exactly three.

Do these criteria extend to higher-order or to systems of ODEs?

Yes, with the same spirit. For an $n$ -th order linear equation $y^{(n)}+p_{n-1}y^{(n-1)}+\cdots+p_0y=0$ , the point $x_0$ is regular singular (Fuchsian) when $(x-x_0)^k p_{n-k}$ is analytic for each $k=1,\dots,n$ — the second-order case $(x-x_0)p_1$ , $(x-x_0)^2p_0$ is just $n=2$ . For first-order systems $\mathbf y'=A(x)\mathbf y$ , the point is regular singular when $A$ has at worst a simple pole, $A=A_{-1}/(x-x_0)+\text{analytic}$ , and the eigenvalues of the residue matrix $A_{-1}$ play the role of the indicial exponents. The dividing line between "convergent generalised series exist" and "only asymptotic series exist" is universal across all these settings.

What is the connection to the Frobenius method and the indicial equation?

Classification is the diagnosis; Frobenius is the treatment. Once you have certified a point as regular singular, the indicial equation $s(s-1)+p_0s+q_0=0$ hands you the two allowed leading exponents, and the Frobenius method builds a full solution $y=(x-x_0)^s\sum a_n(x-x_0)^n$ around each. The classification is what licenses the method: at an ordinary point you would just set $s=0$ (ordinary power series), at a regular singular point the non-zero or non-integer $s$ is essential, and at an irregular singular point the whole Frobenius ansatz collapses because there is no leading power that balances the equation. So the workflow is always: normalise, classify, then read the exponents and grind the recurrence — the subject of the Frobenius and indicial-equation articles this one leads to.