DifferentialEquations
Chapter 6 · Section 6.2

6.2 Solutions About Singular Points: The Frobenius Method

Section 6.1 gave us power series solutions about ordinary points. But some of the most important places in a physical problem are exactly the singular points: the center of a circular drum, the axis of a cylinder, the origin of a spherical wave. There the coefficients of the equation blow up, and the solutions often do too. An ordinary power series cannot describe a solution that behaves like or near the point, so we need a more flexible form.

A familiar example shows what to expect. The Cauchy–Euler equation (Section 4.5)

has as a singular point. Substituting gives , so the solutions are and . Neither is a power series in : one has a fractional power, the other a negative one. The key observation of this section, due to Frobenius, is that for a large and important class of singular points the solutions look like a power of times a power series:

where the exponent need not be a nonnegative integer. Sometimes a logarithm appears as well, and we will see exactly when.

Regular and irregular singular points

As before, write the equation in normalized form .

Definition 6.2.1 (Regular and irregular singular points)

A singular point of is a regular singular point if both

are analytic at . A singular point that is not regular is called irregular.

In words: at a regular singular point, may blow up no worse than and no worse than . This is exactly how the coefficients of a Cauchy–Euler equation behave at , and that is why the Cauchy–Euler idea is the right starting point. At an irregular singular point the solutions can behave in far wilder ways, and the method of this section does not apply.

Multiplying the normalized equation by gives a convenient standard form. Taking (we can always shift to it with ):

So is a regular singular point exactly when the equation can be written this way with and analytic at . When and are constants, this is a Cauchy–Euler equation.

Example 6.2.1

Find and classify the singular points of

Solution
  1. Normalize. The singular points are and .
  2. At . and . Both are rational functions whose denominators do not vanish at , so both are analytic there: is a regular singular point.
  3. At . still has in the denominator, so it is not analytic at . That is enough: is an irregular singular point. (We need not even check .)

The Frobenius series and the indicial equation

Let be a regular singular point, with the equation in standard form and

We look for a solution on of the form

The condition is not a restriction: it simply says that is the lowest power that actually appears. We call such a series a Frobenius series.

Differentiate term by term. Because is in general not an integer, no terms drop out, and all sums still start at :

So

The lowest power. Substitute into . The lowest power of that can appear is , and it comes only from the terms multiplied by the constant terms , :

Its coefficient must vanish, and , so must satisfy

Definition 6.2.2 (Indicial equation)

For with , , the quadratic equation

is the indicial equation; its roots are the exponents (or indicial roots) at the singular point. When both are real we label them so that .

Notice that this is exactly the Cauchy–Euler indicial equation of Section 4.5 for the "frozen" equation . Near the equation behaves like this Cauchy–Euler equation, and the exponents tell us how the solutions behave there: like and , possibly with a logarithm.

The higher powers. Collecting the coefficient of for gives

You do not need to memorize this; in practice we substitute directly, as in the examples. But it shows the crucial point: each is found by dividing by . Everything is fine as long as .

  • For the larger root : would mean that is also a root. But the only roots are and , and . So for all , and the larger root always gives a Frobenius solution.
  • For the smaller root : exactly when , that is, when . So trouble can arise only if is a positive integer, at the step . And if , the two "solutions" coincide, so the second one must be found another way.

This is the whole story behind the famous three cases.

Figure 6.2.1. The three possibilities for real indicial roots. What matters is the difference : not an integer (Case 1), zero (Case 2), or a positive integer (Case 3).

Frobenius' theorem: the three cases

Theorem 6.2.1 (Frobenius)

Let be a regular singular point of , and suppose the Taylor series of and about both converge for , where . Let be the roots of the indicial equation, real, with . Then on there are two linearly independent solutions , where always

and has the following form.

  1. Case 1: is not an integer.
  2. Case 2: .
  3. Case 3: , a positive integer. where the constant may turn out to be zero.

All the power series that appear converge at least for . The general solution on is .

Some remarks that make the theorem easier to use:

  • Finding . For polynomial coefficients, is at least the distance from to the nearest other singular point (complex ones included), just as in Section 6.1. If there is no other singular point, .
  • Negative . On the same statements hold with replaced by and by .
  • Complex exponents. If with , the difference is not an integer, so Case 1 applies with complex coefficients; the real and imaginary parts of give two real solutions (compare the complex case of Cauchy–Euler, Section 4.5).
  • Case 3 with . In Case 3 the step of the recurrence for reads . If the "something" is not zero, the equation is impossible, there is no second Frobenius series, and the logarithm is needed (). If the "something" happens to be zero, can be chosen freely and : we get two Frobenius series after all.
  • Behaviour at . The exponents tell you which solutions stay bounded as . In applications (temperature at the center of a disc, displacement at the center of a drum) we usually keep only the bounded solution.

How to find the logarithmic solution

In Cases 2 and 3 there are two practical routes.

  1. Substitute the form. Put into the equation and solve for and the . The logarithms cancel because is a solution. The following identity, which you can check with the product rule, does most of the work: for , Example 6.2.4 uses this route.
  2. Reduction of order (Section 4.1): , expanding the integrand in a series before integrating. Example 6.2.3 uses this route; it is convenient when has a closed form.

(A third route for Case 2, elegant but more abstract, is to keep as a variable in the series and differentiate with respect to at .)

Method in a nutshell
  1. Check that is a regular singular point; find .
  2. Read off and from the standard form and solve the indicial equation . Decide the case from .
  3. Substitute (all sums from ), shift so every sum contains the same power, and get the recurrence with still a letter. The lowest power reproduces the indicial equation, which is a useful check.
  4. Put in the recurrence and compute (with ). This gives .
  5. Case 1: put and repeat to get . Case 3: try ; if the step is consistent, you get with ; otherwise use the logarithmic form. Case 2: use the logarithmic form.
  6. Write on .

The solver below accepts the standard form with polynomial and . It forms the indicial equation, classifies the case, derives the recurrence and computes the first terms. The first preset is Example 6.2.2 (multiplied by to reach standard form); the second is Example 6.2.4, where the logarithm is unavoidable. Enter your own and to explore the three cases.

Worked examples

Example 6.2.2 (Case 1)

Find two linearly independent Frobenius solutions of near .

Solution
  1. Classify. and , so is singular. But and are analytic: regular singular point. There is no other singular point, so .
  2. Indicial equation. , : The difference is not an integer: Case 1.
  3. Substitute into the original equation (this avoids fractions):
  4. Combine the first two sums. .
  5. Shift the third sum to the power : replace by , so it becomes .
  6. Peel off and combine. The first term gives , the same indicial roots (good check).
  7. Recurrence: for .
  8. . Then , so
  9. . Then :
  10. General solution for : Both series converge for all (the ratio of consecutive coefficients tends to ).

Notice that is an ordinary power series (its exponent is ) while is not: has a vertical tangent at the origin. Both behaviours are visible in Figure 6.2.2.

Figure 6.2.2. The two Frobenius solutions of Example 6.2.2. starts like , with infinite slope at ; starts at with slope . Both are bounded near the origin.
Example 6.2.3 (Case 2: equal roots)

Solve near .

Solution
  1. Classify. Multiplying by gives the standard form , with , . So is a regular singular point and .
  2. Indicial equation. , : , a double root . Case 2.
  3. Substitute. The terms together give . The terms give , which after the shift becomes .
  4. Recurrence. , so for .
  5. First solution. With : , hence and Check: .
  6. Second solution by reduction of order. In normalized form , so and . Then
  7. Expand and integrate. so, term by term,
  8. Multiply by . (for example, the coefficient is ). This is exactly the form promised in Case 2: plus a power series starting at .
  9. General solution for :

As , like , while . So the only solutions that stay bounded at the origin are the multiples of . (For the curious: , where is the exponential integral and is Euler's constant.)

Figure 6.2.3. Example 6.2.3. The second solution inherits the logarithmic singularity of (dashed) at the origin; the first solution is perfectly smooth. Equal indicial roots always produce such a logarithm.
Example 6.2.4 (Case 3: the logarithm is needed)

Find two linearly independent solutions of near . (In Example 6.1.5 we expanded the same equation about the ordinary point ; now we work at the singular point itself.)

Solution
  1. Classify. Standard form: , so , : regular singular point, .
  2. Indicial equation. : , , and . Case 3.
  3. Recurrence. Substituting : the terms give and gives (after shifting) . So
  4. First solution (). : So .
  5. Try . The recurrence becomes . For : , so . For : . Since , this is impossible. There is no second Frobenius series, so .
  6. The logarithmic form. Write with , . For our operator , the product rule gives since . So becomes .
  7. Both sides as series in . On the left, as in step 3 with , On the right, ; with this is .
  8. Match coefficients. For each : (right side zero for ).
  9. The free coefficient. At the coefficient is multiplied by , so it is arbitrary. Changing it only adds a multiple of to (look: is the start of ). We choose . Then and .
  10. Result. and the general solution is for .

Near the origin : it is the exponent that sets the strength of the singularity; the logarithm is a milder correction.

Example 6.2.5 (Case 3 without a logarithm)

Solve near , and identify the solutions in closed form.

Solution
  1. Indicial equation. , , so , : , , : Case 3.
  2. Recurrence. Substituting , the terms give and gives . Hence
  3. . . Since , , and then all odd coefficients vanish. For even : , giving , , :
  4. . Now : The dangerous step reads . It is consistent, is arbitrary, and no logarithm is needed (). Choose (a nonzero choice would only add ).
  5. Continue. gives ; gives :
  6. Closed forms. One can check by direct substitution that are solutions, and expanding and reproduces the two series above. For instance .

Why did the logarithm disappear? The recurrence links to , so odd and even coefficients never interact. The trouble spot is odd, and all odd coefficients were already zero. Whenever the recurrence has step 2 and is odd, Case 3 comes without a logarithm. These two functions are, up to constant factors, the spherical Bessel functions of order 1; they and their combinations describe standing and outgoing waves around a sphere in acoustics and electromagnetics.

Figure 6.2.4. Example 6.2.5. The solution for the larger exponent vanishes at the origin; the one for blows up like . The dashed curve is the three-term Frobenius polynomial for , accurate for small .
Common mistakes
  • Applying the method at an irregular singular point. Always check that and are analytic first.
  • Forgetting that in a Frobenius series no terms drop out on differentiation: the sums for and start at , unlike in Section 6.1.
  • Starting with the smaller root in Case 3. Always build from the larger root; it never fails.
  • Concluding "logarithm needed" in Case 3 without checking the step . As Example 6.2.5 shows, it may be consistent.
  • Reading from the wrong form. They are the constant terms of and in , so first divide or multiply to reach this form.

Exercises

  1. Find and classify the singular points: (a) ; (b) .
    Answer (a) and , both regular. (b) irregular ( is not analytic at ); regular.
  2. Find the indicial roots and decide which case of Theorem 6.2.1 applies (do not solve): (a) ; (b) ; (c) .
    Answer (a) : Case 1. (b) (double): Case 2. (c) : Case 3 with .
  3. Find two Frobenius solutions of (terms through inside the brackets).
    Answer ; , . (In fact , .)
  4. Find two Frobenius solutions of .
    Answer ; , .
  5. Show that solves , and find the second solution in the form (terms through ).
    Answer (double); .
  6. For , find the Frobenius solution (general term) and write down the form of a second solution.
    Answer (double); ; .
  7. For : find , show that the smaller root does not give a Frobenius series, and find and the first terms of (take , ).
    Answer ; ; for the step gives , impossible. ,
  8. Solve . Show that Case 3 occurs but no logarithm is needed, and find both solutions in closed form.
    Answer (, recurrence with step 2); .
  9. Explain why the Frobenius method cannot be used at for .
    Answer , so is not analytic at : the singular point is irregular.
  10. Without solving, give an interval on which the Frobenius series about of are guaranteed to converge.
    Answer , analytic for (singular points ): the series converge on (and on with ).