Frobenius method — regular singular points: Visual walkthrough

Before we begin, the promise: by the end you will see why the exponent $r$ is not chosen but forced by the equation, and why only the very lowest power of $x$ matters for finding it.

Step 1 — Why a plain power series cannot even start here

WHAT. We look at an equation whose coefficients blow up at $x = 0$ , and we watch a plain Taylor series fail.

Write any second-order linear ODE in standard form (divide until $y^{''}$ stands alone): $curvature y^{''} + friction coeff. P (x) slope y^{'} + spring coeff. Q (x) height y = 0.$

$y$ is the unknown function (think: height of a swing over time $x$ ).
$y^{'}$ is its slope, $y^{''}$ its curvature.
$P (x), Q (x)$ are known functions — the "coefficients". At an ordinary point they are finite; here they explode as $x \to 0$ .

WHY. A plain power series $y = \sum a_{n} x^{n}$ is smooth — it has a finite value and finite slope at $x = 0$ . But if $P$ or $Q$ shoots to infinity there, the true solution may itself misbehave (like $x^{1/2}$ or $x^{- 1}$ ), which no ordinary Taylor series can represent. So the plain method is doomed before it starts.

PICTURE. The blue curve $x^{1/2}$ has a vertical tangent at $x = 0$ — infinite slope. No polynomial-like Taylor curve (amber) can match that.

Figure — Frobenius method — regular singular points

Contrast this with an ordinary point, where a clean Taylor series does work.

Step 2 — Taming the coefficients: the RSP test

WHAT. We check that the blow-up is mild enough to fix, by multiplying $P$ and $Q$ by exactly the right powers of $x$ .

Define the tamed coefficients: $p (x) = one factor x P (x), q (x) = two factors x^{2} Q (x) .$

$p (x) = x P (x)$ : one power of $x$ cancels a $1/ x$ blow-up in $P$ .
$q (x) = x^{2} Q (x)$ : two powers cancel a $1/ x^{2}$ blow-up in $Q$ .

If both $p$ and $q$ come out analytic (finite, Taylor-expandable) at $x = 0$ , then $x = 0$ is a regular singular point (RSP) and Frobenius is guaranteed to work.

WHY these exact powers? Because the strongest blow-up a single factor $x^{r}$ can balance in the equation $x^{2} y^{''} + x p y^{'} + q y = 0$ is $P \sim 1/ x$ and $Q \sim 1/ x^{2}$ . Weaker or equal — we cope. Stronger — the point is irregular and the method breaks.

PICTURE. $P (x) = 3/ (2 x)$ (cyan) rockets to infinity; multiply by $x$ and the amber curve $x P$ flattens to a finite value $\frac{3}{2}$ at $x = 0$ . That finite landing value is what we will call $p_{0}$ .

Analyticity here means "has a genuine radius of convergence".

Step 3 — The Frobenius ansatz: steal the Taylor idea, add a magic power

WHAT. We guess a solution shape that is a Taylor series multiplied by an unknown power $x^{r}$ .

$y = \sum_{n = 0}^{\infty} a_{n} x^{n + r}, a_{0} \neq = 0.$

$x^{r}$ = the magic leading power, $r$ unknown (could be $\frac{1}{2}$ , $- 1$ , anything).
$\sum a_{n} x^{n}$ = the ordinary smooth Taylor part riding on top.
$a_{0} \neq = 0$ : the very first coefficient is nonzero, so $x^{r}$ is genuinely the lowest power present.

WHY $a_{0} \neq = 0$ ? If $a_{0} = 0$ , the series would really start at $x^{r + 1}$ — but then we should have called that the leading power and renamed $r$ . Insisting $a_{0} \neq = 0$ pins $r$ to the true bottom of the series, with no ambiguity.

PICTURE. The amber envelope $x^{r}$ sets the overall shape near $0$ ; the cyan smooth wiggle $\sum a_{n} x^{n}$ decorates it. The product (white) is our candidate $y$ .

Step 4 — Differentiate the ansatz term by term

WHAT. We compute $y^{'}$ and $y^{''}$ so we can substitute into the ODE.

Bring down the exponent each time (that is what differentiating a power does): $y^{'} = \sum_{n} a_{n} old exponent (n + r) x^{n + r - 1},$ $y^{''} = \sum_{n} a_{n} two exponents pulled down (n + r) (n + r - 1) x^{n + r - 2} .$

Each differentiation multiplies term $n$ by its current exponent and lowers that exponent by $1$ .
So $y^{''}$ carries the product $(n + r) (n + r - 1)$ — the signature of "differentiated twice".

WHY. The ODE is written in terms of $y, y^{'}, y^{''}$ . To turn it into equations for the unknowns $a_{n}$ and $r$ , we must express all three as series in the same variable $x$ .

PICTURE. A "staircase" of exponents: $y$ sits at height $n + r$ , $y^{'}$ one step down at $n + r - 1$ , $y^{''}$ two steps down at $n + r - 2$ . The multipliers appear on the arrows.

Step 5 — Multiply by $x^{2}$ and read off the LOWEST power only

WHAT. We put the ODE in the clean form $x^{2} y^{''} + x p (x) y^{'} + q (x) y = 0$ and collect only the smallest power of $x$ , which is $x^{r}$ (the $n = 0$ term).

Multiplying the derivatives by $x^{2}$ and $x$ lifts every exponent back up to $n + r$ : $x^{2} y^{''} \to a_{0} (r) (r - 1) x^{r}, x p y^{'} \to a_{0} p_{0} r x^{r}, q y \to a_{0} q_{0} x^{r},$ where $p_{0} = p (0) = lim_{x \to 0} x P$ and $q_{0} = q (0) = lim_{x \to 0} x^{2} Q$ are the landing values from Step 2 (only the constant term of each tamed series survives at the lowest power).

$r (r - 1)$ = contribution of $x^{2} y^{''}$ at $n = 0$ .
$p_{0} r$ = contribution of $x p y^{'}$ at $n = 0$ .
$q_{0}$ = contribution of $q y$ at $n = 0$ .

WHY the lowest power only? Every power of $x$ must vanish separately for the whole sum to be zero. The lowest one, $x^{r}$ , is the only place where $r$ appears alone (no $a_{1}, a_{2}, \dots$ mixed in). So it hands us a clean equation for $r$ .

PICTURE. A stacked-coefficient tower: the $x^{r}$ floor collects three pieces $r (r - 1)$ , $p_{0} r$ , $q_{0}$ ; the higher floors ( $x^{r + 1}, \dots$ ) involve later $a_{n}$ and are set aside.

Step 6 — The indicial equation appears

WHAT. Set the coefficient of $x^{r}$ to zero. Since $a_{0} \neq = 0$ , we may divide it out.

$a_{0} [r (r - 1) + p_{0} r + q_{0}] = 0 ⟹ r (r - 1) + p_{0} r + q_{0} = 0$

$F (r)$ = the indicial polynomial — a plain quadratic in $r$ .
Its roots are the allowed magic powers; nothing else can lead the series.

WHY it's a quadratic. A second-order ODE has two independent solutions. A quadratic has two roots. The one balances the other: two exponents for two behaviours near $x = 0$ .

PICTURE. The parabola $F (r)$ crosses the horizontal axis at $r_{1}$ and $r_{2}$ . Those two crossings are the only exponents the equation permits.

Compare the Euler–Cauchy equation, whose characteristic equation is exactly this indicial equation with $p_{0}, q_{0}$ constant — the RSP case is its curved generalisation.

Step 7 — Edge cases: the root difference decides everything

WHAT. We classify by $r_{1} - r_{2}$ , because that difference controls whether the two series collide.

The higher coefficients obey the recurrence (coefficient of $x^{n + r}$ ): $F (n + r) a_{n} = - \sum_{k = 1}^{n} [p_{k} (n - k + r) + q_{k}] a_{n - k} .$ Trouble strikes only when $F (n + r) = 0$ — i.e. when $n + r_{2}$ lands on the other root $r_{1}$ .

Case 1 · $r_{1} - r_{2} \in / Z$ : $F (n + r_{2})$ never hits zero → two clean independent series. Check they are independent via the Wronskian.
Case 2 · $r_{1} = r_{2}$ : only one exponent exists; the second solution is forced to carry $ln x$ .
Case 3 · $r_{1} - r_{2} = N \in Z^{+}$ : at step $n = N$ , $F (r_{2} + N) = F (r_{1}) = 0$ → possible division by zero → a $ln x$ may appear (coefficient $κ$ can be $0$ ).

WHY the log? When $F (n + r_{2}) = 0$ the recurrence demands dividing by zero — the power series runs out of room for a second solution. Reduction of order with $y_{2} = v y_{1}$ then produces $\int \frac{d x}{x} = ln x$ : the mathematics supplies the missing independent solution by hand.

PICTURE. Three number lines of exponents. Case 1: the two combs of powers interleave, never overlap. Case 2: they coincide (one comb). Case 3: they collide at step $N$ (marked amber) — the danger point.

The equal-root Case 2 is exactly what happens in Bessel's equation of order 0 ( $r_{1} = r_{2} = 0$ ), giving $J_{0}$ and a log-carrying $Y_{0}$ . Legendre's equation sits at an ordinary point instead, so no $x^{r}$ is needed there.

The one-picture summary

This single blueprint compresses the pipeline: blow-up → tame with $x$ and $x^{2}$ → RSP? → ansatz $x^{r} \sum a_{n} x^{n}$ → lowest power gives $F (r) = 0$ → root difference picks the case.

Recall Feynman retelling of the whole walkthrough

A swing's chain is frayed right at the pivot $x = 0$ , so the coefficients $P, Q$ scream to infinity there (Step 1). We test how bad the fraying is: multiply $P$ by $x$ and $Q$ by $x^{2}$ ; if both settle to finite numbers $p_{0}, q_{0}$ , the fraying is tame — a regular singular point (Step 2). Now we cheat cleverly: guess the motion looks like $x^{r}$ (a magic power) times an ordinary smooth wiggle, with the first coefficient not zero so $x^{r}$ is truly the bottom (Step 3). We differentiate this guess (Step 4), plug it into the tidy $x^{2} y^{''} + x p y^{'} + q y = 0$ , and stare only at the smallest power $x^{r}$ , because that's the one spot where $r$ stands alone (Step 5). Setting its coefficient to zero drops out a tiny quadratic $r (r - 1) + p_{0} r + q_{0} = 0$ — the indicial equation (Step 6). Its two roots are the only allowed magic powers. Finally their difference tells the story: far apart (non-integer) → two clean series; equal or a whole-number gap → the two solutions try to be the same, the recurrence divides by zero, and a $ln x$ sneaks in to rescue a genuinely different second solution (Step 7).

Active Recall

Recall Self-test on the derivation

Which single power of $x$ gives the indicial equation, and why? ::: The lowest, $x^{r}$ (the $n = 0$ term), because there $r$ appears with no other $a_{n}$ mixed in. What are $p_{0}$ and $q_{0}$ as limits? ::: $p_{0} = lim_{x \to 0} x P (x)$ , $q_{0} = lim_{x \to 0} x^{2} Q (x)$ . Why must $a_{0} \neq = 0$ ? ::: So $x^{r}$ is genuinely the lowest power; else we'd rename $r$ . What exactly triggers the $ln x$ in the second solution? ::: $F (n + r_{2}) = 0$ at some step $n$ , making the recurrence divide by zero.