JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Singular Points and the Method of Frobenius

A plain power series works beautifully — until the leading coefficient vanishes and the whole machine stalls. The method of Frobenius repairs it with one daring change: allow the series to start at a fractional or negative power.

Where the plain power series breaks

In the first two guides of this rung you built solutions of a second-order linear equation as a clean power series y = sum a_n x^n centred on a point, then chased a recurrence relation for the coefficients. That whole programme rested on a quiet assumption: that the point you expanded around was an ordinary point, a place where the equation is well-behaved. This guide asks what happens at the OTHER kind of point — and how to rescue the series method when it is exactly the interesting point you cannot avoid.

Write the equation in standard form, P(x) y'' + Q(x) y' + R(x) y = 0, and put a point x0 under the microscope. The trouble starts the moment the leading coefficient P(x0) = 0. Dividing through to reach the normal form y'' + p(x) y' + q(x) y = 0 forces a division by zero, so p(x) = Q/P and q(x) = R/P can blow up at x0. Such a point is a singular point: the equation's normalised coefficients fail to be analytic there. At a singular point the comfortable theorem that guarantees a power-series solution simply does not apply.

Two flavours of singular: regular and irregular

Not all singular points are equally savage. The decisive question is how BADLY p and q blow up as you approach x0. If they merely have a 'mild' singularity — p with at most a simple pole and q with at most a double pole — the explosion is gentle enough to tame. The precise test multiplies the offending coefficients by just enough of (x - x0) to cancel the blow-up: form (x - x0) p(x) and (x - x0)^2 q(x), and ask whether BOTH of these are analytic (well-behaved, expandable in a Taylor series) at x0.

If both products pass that test, x0 is a regular singular point — singular, yes, but tame, and the method of this guide will work there. If either product still misbehaves, x0 is an irregular singular point, a genuinely wild place where even Frobenius gives up and you need heavier machinery (asymptotic expansions, for example). The whole sorting rule fits in one short table, and learning to apply it in ten seconds is the practical skill of this guide.

  classify x0 in   P(x) y'' + Q(x) y' + R(x) y = 0

  p(x) = Q/P,   q(x) = R/P

  P(x0) != 0 ........................ ORDINARY point  -> plain power series
  P(x0) =  0  and  both
     (x - x0)*p(x)   analytic at x0
     (x - x0)^2*q(x) analytic at x0 .. REGULAR singular -> Frobenius works
  otherwise ......................... IRREGULAR singular -> Frobenius fails

  example:  x^2 y'' + x y' + (x^2 - v^2) y = 0   (Bessel, at x0 = 0)
     p = 1/x       -> x*p   = 1        analytic   OK
     q = (x^2-v^2)/x^2 -> x^2*q = x^2 - v^2  analytic   OK
     => x = 0 is a REGULAR singular point
The classification test in full, with Bessel's equation worked through: its only finite singular point, x = 0, is regular — which is exactly why Frobenius can reach the Bessel functions.

Why a plain series cannot work — and the one repair

To feel why the ordinary recipe fails, try the simplest singular equation you can imagine: the Cauchy-Euler equation x^2 y'' - 2 y = 0. Its solutions are y = x^2 and y = x^(-1). Look at that second one. A plain power series y = a_0 + a_1 x + a_2 x^2 + ... is built only from non-negative integer powers, so it can NEVER reproduce x^(-1) — the answer literally lives outside the space the method searches. Forcing the recurrence anyway collapses to nonsense, because the leading-coefficient bookkeeping divides by zero at the very first step.

Frobenius's idea is disarmingly simple: if the answer wants a power like x^(-1) or x^(1/2), then LET the series start at a power we do not fix in advance. Multiply an ordinary power series by an unknown factor x^r, where the exponent r is a number to be determined. The trial solution becomes y = x^r * sum a_n x^n = sum a_n x^(n+r), with a_0 not zero. The single new unknown r is the whole repair — it lets the series begin at a fractional, negative, or otherwise non-integer power exactly when the equation demands it.

Running the method: the lowest power decides r

The mechanics of the method of Frobenius mirror the plain power-series method you already know, with one new opening move. You substitute y = sum a_n x^(n+r) and its derivatives into the equation, then collect terms by power of x. Because every term now carries the offset r, the lowest power of x that appears is x^r itself, and its coefficient must vanish like every other. That single lowest-power condition is special: it involves only a_0 and r, and since a_0 is not zero, it becomes an equation for r alone.

That equation for r is the indicial equation — a quadratic in r whose two roots r1 and r2 are the admissible starting exponents, the 'indices' the solution may take off from. The remaining powers of x give the usual recurrence relation, but now the recurrence carries r as a parameter, so each choice of root generates its own coefficient chain. The walkthrough below shows the spine of the calculation; the careful study of WHAT the two roots tell you is the subject of the very next guide.

  1. Confirm x0 is a regular singular point with the (x - x0) p and (x - x0)^2 q test; only then is Frobenius licensed.
  2. Substitute the Frobenius series y = sum a_n x^(n+r) and its term-by-term derivatives into the equation.
  3. Collect by powers of x; set the coefficient of the LOWEST power (the a_0 term) to zero to read off the indicial equation in r.
  4. Solve the indicial equation for its two roots r1 and r2 (take r1 >= r2 when they are real).
  5. Feed a root back into the general-power condition to get the recurrence, then grind out a_1, a_2, ... to build that root's series solution.

What the method promises, and where it stays silent

Be honest about the guarantee. Frobenius promises that the LARGER root r1 always yields a genuine series solution of the form x^r1 times an analytic series, valid on the same kind of disc as before — its radius of convergence reaches at least to the nearest OTHER singular point of the equation. The smaller root r2, however, is delicate: it produces a clean second solution only in the lucky case. When r1 - r2 is a non-negative integer, the two roots can collide or interfere, and a logarithm term log(x) can be forced into the second solution. Those three outcomes — distinct roots, repeated root, integer-difference root — are exactly the three cases catalogued next.

It is worth pausing on the most common misconception. People assume that because we expanded 'in a series', the solution must be a nice analytic function with no surprises at x0. Not so: the factor x^r often makes the solution non-analytic right at the singular point — x^(1/2) has a vertical tangent there, x^(-1) blows up, and a log term diverges. The series captures the solution NEAR x0, but the singularity of the equation leaves a fingerprint on the solution itself. That fingerprint is information: it is how the centre of a drum or the origin of an atom announces its special status in the mathematics.

Finally, remember the boundary of the whole approach. Frobenius is licensed only at a REGULAR singular point; at an irregular singular point the indicial machinery can fail to give two usable roots, and the natural expansions there are not power-like at all but exponential and asymptotic. So the method is powerful but not universal — it is precisely the right tool for the equations of mathematical physics, which were practically designed to have regular singular points, and it is the bridge that carries you to the Bessel and Legendre equations of the final guide in this rung.