Differential Equations › Section F · Series solutions › Topic F.4

The Frobenius method

At many singular points the solutions behave like a power of x, as Euler equations showed, but with a power series attached. The Frobenius method finds that power and that series, and it is the standard route to the series for Bessel functions.

A power times a series

At an ordinary point a plain power series always works. At a singular point it can fail: the Euler equation \(2x^2y'' + 3xy' - y = 0\) has solutions \(\sqrt{x}\) and \(x^{-1}\), and neither is a power series in \(x\). The topic Euler (Cauchy-Euler) equations solved that family exactly with \(y = x^r\). For equations that are only Euler-like near the singular point, the German mathematician Georg Frobenius, in the 1870s, combined the two ideas: try a power \(x^r\) multiplied by a power series. The method works at singular points of one particular kind, so the first job is to classify them.

Regular and irregular singular points

Write the equation as \(y'' + P(x)y' + Q(x)y = 0\) and suppose \(x = 0\) is a singular point. It is a regular singular point if \(xP(x)\) and \(x^2Q(x)\) are both analytic at 0; otherwise it is irregular. Regular means the singularity is no worse than an Euler equation's. Record the limits

p_0 = \lim_{x \to 0} xP(x), \qquad q_0 = \lim_{x \to 0} x^2Q(x)

For an Euler equation \(ax^2y'' + bxy' + cy = 0\), \(xP = \frac{b}{a}\) and \(x^2Q = \frac{c}{a}\), so 0 is regular. For \(x^3y'' + y = 0\), \(x^2Q = \frac{1}{x}\), so 0 is irregular. A singular point at \(x_0 \ne 0\) is tested the same way with \(x - x_0\) in place of \(x\).

The Frobenius form

Near a regular singular point at 0, look for solutions on \(x > 0\) of the form

y = x^r \sum_{n=0}^{\infty} c_n x^n = \sum_{n=0}^{\infty} c_n x^{n+r}, \qquad c_0 \ne 0

The exponent \(r\) need not be a whole number, and the condition \(c_0 \ne 0\) makes \(x^r\) the true leading behaviour. Differentiating term by term,

y' = \sum_{n=0}^{\infty} (n+r)c_n x^{n+r-1}, \qquad y'' = \sum_{n=0}^{\infty} (n+r)(n+r-1)c_n x^{n+r-2}

Unlike ordinary power series, these sums still start at \(n = 0\): with \(r\) unknown, no early terms are guaranteed to vanish.

The indicial equation

Substitute the Frobenius form and collect the lowest power of \(x\). Near 0 the equation behaves like the Euler equation \(x^2y'' + p_0xy' + q_0y = 0\), so that lowest coefficient is \(c_0\,[r(r-1) + p_0 r + q_0]\). Because \(c_0 \ne 0\),

r(r-1) + p_0 r + q_0 = 0

This is the indicial equation; its roots, labelled \(r_1 \ge r_2\) when real, are the possible leading exponents. Every higher power of \(x\) then gives a recurrence for \(c_n\), solved once for each root. The resulting series converge at least for \(0 < x < \rho\), where \(\rho\) is the distance from 0 to the nearest other singular point, complex ones included.

Worked example: setting up 2xy″ + y′ + y = 0

1. Classify 0: \(P = Q = \frac{1}{2x}\), so \(xP = \frac{1}{2}\) and \(x^2Q = \frac{x}{2}\), both analytic. The point is regular, with \(p_0 = \frac{1}{2}\) and \(q_0 = 0\). 2. Indicial equation: \(r(r-1) + \frac{r}{2} = 0\), which factors as \(r(r - \frac{1}{2}) = 0\), so \(r_1 = \frac{1}{2}\) and \(r_2 = 0\). Their difference is not a whole number, so two Frobenius series are expected. 3. Substitute: \(2xy''\) and \(y'\) both carry \(x^{n+r-1}\), with combined coefficient \((n+r)(2n+2r-1)c_n\); shifting the sum for \(y\) up by one gives \(c_{n-1}\) at the same power. 4. Setting each coefficient to zero:

c_n = -\frac{c_{n-1}}{(n+r)(2n+2r-1)}, \quad n \ge 1

Worked example: the two solutions

Continue with \(c_0 = 1\). 1. For \(r = 0\) the recurrence is \(c_n = -\frac{c_{n-1}}{n(2n-1)}\): \(c_1 = -1\), \(c_2 = \frac{1}{6}\), \(c_3 = -\frac{1}{90}\). 2. For \(r = \frac{1}{2}\) it is \(c_n = -\frac{c_{n-1}}{n(2n+1)}\): \(c_1 = -\frac{1}{3}\), \(c_2 = \frac{1}{30}\), \(c_3 = -\frac{1}{630}\). 3. So

y_1 = 1 - x + \frac{x^2}{6} - \frac{x^3}{90} + \cdots, \qquad y_2 = \sqrt{x}\Big(1 - \frac{x}{3} + \frac{x^2}{30} - \cdots\Big)

4. Check: these are the Maclaurin patterns of \(\cos\sqrt{2x}\) and of \(\sin\sqrt{2x}\) divided by \(\sqrt{2}\), and direct substitution confirms both solve the equation. There is no other singular point, so both series converge for every \(x > 0\).

Three cases for the roots

The roots decide the form of the second solution, mirroring the Euler cases. The larger root \(r_1\) always gives a Frobenius series \(y_1\). If \(r_1 - r_2\) is not a whole number, \(r_2\) gives a second, independent Frobenius series. If \(r_1 = r_2\), the second solution contains a logarithm:

y_2 = y_1 \ln x + \sum_{n=1}^{\infty} b_n x^{n+r_1}

If \(r_1 - r_2\) is a positive whole number, then

y_2 = k\,y_1 \ln x + x^{r_2}\sum_{n=0}^{\infty} b_n x^n, \quad b_0 \ne 0

where the constant \(k\) may or may not be zero. You find out which by trying the smaller root: if its recurrence breaks down, a logarithm is needed.

Worked example: roots one apart, no logarithm

Consider

x^2y'' + xy' + \Big(x^2 - \frac{1}{4}\Big)y = 0

1. Here \(xP = 1\) and \(x^2Q = x^2 - \frac{1}{4}\), so \(p_0 = 1\), \(q_0 = -\frac{1}{4}\), and the indicial equation \(r^2 - \frac{1}{4} = 0\) gives \(r = \pm\frac{1}{2}\), one apart. 2. Substituting, the coefficient of \(x^{n+r}\) is

\Big[(n+r)^2 - \frac{1}{4}\Big]c_n + c_{n-2}

with \(c_{-1} = 0\). 3. Try the smaller root \(r = -\frac{1}{2}\). At \(n = 1\) the bracket is \(\frac{1}{4} - \frac{1}{4} = 0\), so the equation reads \(0 \cdot c_1 = 0\): no contradiction, and \(c_1\) is simply free. 4. For \(n \ge 2\), \(c_n = -\frac{c_{n-2}}{n(n-1)}\), the cosine and sine patterns. So

y = x^{-1/2}(c_0\cos x + c_1\sin x)

and no logarithm is needed: the integer case really can go either way.

Bessel's equation

That example is the case \(\nu = \frac{1}{2}\) of Bessel's equation,

x^2y'' + xy' + (x^2 - \nu^2)y = 0

which has a regular singular point at 0 with indicial roots \(\pm\nu\). For \(\nu = 0\) the root 0 is repeated; the Frobenius series is

J_0(x) = \sum_{k=0}^{\infty} \frac{(-1)^k}{(k!)^2}\Big(\frac{x}{2}\Big)^{2k} = 1 - \frac{x^2}{4} + \frac{x^4}{64} - \cdots

and the second solution, \(Y_0\), contains \(\ln x\) and is unbounded as \(x \to 0^+\). Bessel functions describe vibrating drumheads, heat flow in cylinders and electromagnetic waves in cables; the topic Special functions in applications looks at them more closely.

Common mistakes to avoid

First, applying the method at an irregular singular point, where it can fail. Second, finding \(p_0\) and \(q_0\) from the original coefficients: divide by the coefficient of \(y''\) first, then multiply by \(x\) and \(x^2\). In the first worked example, \(p_0\) is \(\frac{1}{2}\), not 1. Third, dropping the condition \(c_0 \ne 0\), which is what makes the indicial equation hold. Fourth, starting the derivative sums at \(n = 1\) or \(n = 2\) out of habit; with \(x^{n+r}\) no terms vanish. Fifth, assuming roots a whole number apart always force a logarithm, or never do: test the smaller root.

Key takeaways

A singular point at 0 is regular when \(xP\) and \(x^2Q\) are analytic; there, try \(y = \sum c_n x^{n+r}\) with \(c_0 \ne 0\). The lowest power gives the indicial equation \(r(r-1) + p_0r + q_0 = 0\), and higher powers give a recurrence for each root. Roots not a whole number apart give two Frobenius series; equal roots force \(y_1 \ln x\) in the second solution; roots a positive whole number apart may or may not, so test the smaller root. Bessel's equation, with roots \(\pm\nu\), is the central example.

Flashcards

Flashcard

Regular singular point at 0

xP(x) and x²Q(x) are both analytic at 0

The singularity is no worse than an Euler equation's.

Test your understanding

Question 1 of 5

Which statement classifies x = 0 for x³y″ + y = 0?

Question 2 of 5

What is the indicial equation of 2xy″ + y′ + y = 0 at x = 0?

Question 3 of 5

The indicial roots at a regular singular point are equal. What does the second solution contain?

Question 4 of 5

For x²y″ + xy′ + (x² - 1/4)y = 0, what are the indicial roots and is a logarithm needed?

Question 5 of 5

In y = Σ cₙxⁿ⁺ʳ, why is c₀ required to be non-zero?

← Previous topicNext topic →