The Exponent Comes Before the Coefficients
In the previous guide you set up the [[method-of-frobenius|method of Frobenius]] at a regular singular point: instead of a plain power series, you try a solution of the form y = x^r (a_0 + a_1 x + a_2 x^2 + ...), where the unknown leading exponent r is allowed to be any number — negative, fractional, anything. That extra factor x^r is what lets the series survive next to a singularity, where a bare power series from an ordinary point would have no chance. Now we ask the obvious next question: *what is r actually allowed to be?*
Here is the small miracle. When you substitute y = x^r (a_0 + ...) into the equation and collect powers of x, the very lowest power that appears does not mix any of the higher coefficients in at all — it is just a_0 times a polynomial in r. Because a_0 is by definition nonzero (it is the very first nonzero coefficient), that polynomial itself must be zero. So before you compute a single a_1, a_2, ..., the equation forces a clean algebraic condition on r alone. That condition is the [[ode-indicial-equation|indicial equation]], and it is the gatekeeper: only the r-values it permits can ever start a valid Frobenius solution.
Reading Off the Indicial Equation
There is a quick recipe that saves you from re-deriving it every time. Write the equation in the standard regular-singular form x^2 y'' + x p(x) y' + q(x) y = 0, where p(x) and q(x) are ordinary analytic functions near x = 0. Let p_0 = p(0) and q_0 = q(0) — just the *constant terms* of those two functions. Then the indicial equation is exactly r(r - 1) + p_0 r + q_0 = 0. That is it: a simple quadratic. The honest and slightly surprising fact is that only p_0 and q_0 enter — the rest of the Taylor expansions of p and q do not touch the exponents at all. They will matter later, for the coefficients; the *leading behaviour* is decided by two numbers.
Put the ODE in standard form:
x^2 y'' + x p(x) y' + q(x) y = 0 ( p, q analytic at x = 0 )
Read the constant terms:
p_0 = p(0) q_0 = q(0)
Indicial equation (a quadratic in r):
r(r - 1) + p_0 r + q_0 = 0
Example - Bessel's equation of order nu:
x^2 y'' + x y' + (x^2 - nu^2) y = 0
p_0 = 1 , q_0 = -nu^2
r(r-1) + r - nu^2 = r^2 - nu^2 = 0 -> r = +nu , r = -nuTake a moment to feel what the two roots r_1 and r_2 *mean*. They are the leading exponents of the two solutions near the singular point, so they forecast the local shape of each: a positive r means the solution starts off small and well-behaved, a negative r means it blows up like 1/x^|r|, and a fractional r means it picks up a branch like sqrt(x). For Bessel's equation the roots came out as +nu and -nu, which are exactly the orders of the two Bessel functions you will meet in the final guide. Reading the indicial roots first is like reading the first word of each solution before writing the rest of the sentence.
The Whole Plot Hinges on One Number: r_1 - r_2
Order the roots so that r_1 >= r_2 (when they are real). The single quantity that decides the rest of the story is the difference r_1 - r_2. This is the heart of the [[frobenius-three-cases|three cases of the Frobenius roots]]. The larger root r_1 *always* yields one clean Frobenius series — you can rely on that no matter what. The whole question, the only question, is what the *second*, independent solution looks like: does the smaller root r_2 hand you a second clean series, or does the machinery jam and force a logarithm to appear? The gap r_1 - r_2 answers exactly that.
Why should the *difference* be the deciding number, rather than the roots themselves? The intuition is sharp and worth holding onto. As you build the second solution from the smaller root r_2, the recurrence keeps dividing by quantities of the form (r_2 + n)(r_2 + n - 1) + p_0(r_2 + n) + q_0 — which is the indicial polynomial evaluated at r_2 + n. That expression is zero exactly when r_2 + n equals the *other* root r_1, i.e. when n = r_1 - r_2 is a positive integer. A zero in the denominator is the machine jamming. So the trouble is not about r_1 or r_2 alone; it is about whether stepping up from r_2 by whole numbers ever lands you back on r_1. The difference is the whole answer.
The Three Cases, Walked Through
Now the taxonomy almost writes itself. Case 1 — the difference r_1 - r_2 is not an integer (this includes fractions like 1/2 or irrationals). This is the lucky, easy case: stepping up from r_2 by whole numbers never hits r_1, so no denominator ever vanishes. Each root delivers its own independent series, y_1 = x^(r_1)(series) and y_2 = x^(r_2)(series), and you simply combine them. Complex conjugate roots fall here too, by the way — their difference is purely imaginary, never a positive integer.
Case 2 — the roots are equal, r_1 = r_2. A double root can only ever hand you *one* series, so a second, independent solution must come from somewhere else — and it always arrives carrying a logarithm: y_2 = y_1 ln(x) + x^(r_1)(a new series). The logarithm is not optional here; it is forced, exactly as the repeated-root case for constant-coefficient and Cauchy-Euler equations forced an extra factor of t or ln x. If you stubbornly look for two plain series, you will only ever find one and stall with half a general solution.
Case 3 — the roots differ by a positive integer, r_1 - r_2 = N. This is the genuinely subtle one, and the only honest thing to say is: *it depends, and you have to check.* The larger root r_1 still gives its clean series. The smaller root r_2 *might* give a second independent series — or the recurrence might hit that fatal zero at step N and force the form y_2 = c y_1 ln(x) + x^(r_2)(series). The constant c can turn out to be zero (then no logarithm, and you got lucky) or nonzero (then a genuine logarithm). You cannot tell which from the difference alone; you must actually compute c. Beware the tempting half-truth that an integer difference *always* means a logarithm — it does not.
Where the Logarithm Comes From, and Why It Matters
When a case does force a logarithm, you do not find it by accident — you build it in deliberately. Two standard routes get you there. One is reduction of order: take the known solution y_1 and look for a second solution of the form y_2 = u(x) y_1; substituting reduces the problem to a first-order equation for u', and an integral of 1/y_1^2 quietly produces the ln(x). The other route is to treat r as a continuous variable, differentiate the series solution with respect to r, and evaluate at the repeated root — differentiating x^r with respect to r gives x^r ln x, which is precisely where the logarithm is born. Here is a step-by-step view of how the whole case-analysis plays out in practice.
- Confirm x = 0 is a regular singular point (the previous guide's classification), then write the equation in standard form x^2 y'' + x p(x) y' + q(x) y = 0.
- Read off p_0 = p(0), q_0 = q(0) and form the indicial equation r(r-1) + p_0 r + q_0 = 0. Solve for the two roots r_1 >= r_2.
- Classify by r_1 - r_2: not an integer (Case 1), zero (Case 2), or a positive integer (Case 3). This single number tells you the form of the second solution before any coefficients.
- Build y_1 from the larger root r_1 using the recurrence relation from earlier in this rung — this series always works and always converges out to the next singular point.
- Build y_2 by the rule of the case: a second clean series (Case 1), a forced y_1 ln(x) + series (Case 2), or compute the constant c to decide whether the log is present (Case 3).
Why bother with all this care? Because the indicial roots and their three cases are not bookkeeping — they are the structural skeleton of the special functions waiting in the next guide. The logarithmic second solution of Case 2 is exactly how Bessel functions of integer order acquire their 'second kind' partners; the indicial roots +nu and -nu are why Bessel functions come in pairs. And a final honest reminder that carries from the whole rung: even when Frobenius succeeds, the series it produces converges only out to the nearest *other* singular point — its radius of convergence is set by the geometry of the singularities, not by your effort. The method is powerful, but it never promises a closed form, only a convergent local description.