At a regular singular point, the indicial roots suggest possible leading powers but do not guarantee two independent power-series solutions. Bessel's equation makes the issue concrete. For noninteger order we can construct two solutions and prove their convergence directly. At integer order a recurrence can fail; at order zero a second solution contains a logarithm.
We work on with real order :
The point is regular singular, not an ordinary point of the equation after division by . Our normalization below sets the first Frobenius coefficient to one; standard Bessel functions may use different constants.
The recurrence and convergence
A trial gives the indicial equation . With , coefficient matching gives
The even-coefficient construction chooses all odd coefficients to be zero; this is a choice when the recurrence leaves an odd coefficient free. For and for when is not a positive integer, define
where the empty product is one. All denominators in the stated cases are nonzero.
Theorem 2.5.1
(Convergent Bessel Frobenius solutions)
For in the stated cases, is a real solution of (1) on . The series in has infinite radius of convergence, and its derivatives can be taken term by term on compact subsets of .
If is not an integer, and are linearly independent and
Every real solution of (1) on is their linear combination.
Proof
(Prove the series before using the Wronskian)
For an even coefficient, (2) with becomes
Starting with gives the displayed product. For fixed , the ratio of successive absolute even terms is
Thus the power series in has infinite radius. The standard termwise-differentiation theorem for power series gives derivative series with the same radius of convergence and uniform convergence on compact sets; multiplying by and differentiating is legitimate on compact subsets of . The recurrence then proves (1) there.
As , and . For noninteger, these give . Dividing (1) by and differentiating the Wronskian gives , so is constant and . It never vanishes. The two solutions therefore form a basis of the two-dimensional solution space on by the initial-value uniqueness theorem at any positive point.
Proof X-ray
Formal recurrence versus an actual solution
The proof has three separate steps, and each uses a different hypothesis. The recurrence only produces coefficients; the case restriction on is used exactly once, to keep every factor in the denominators nonzero. The ratio test then turns coefficients into a function: the ratio tends to zero for every because the denominator grows like , so the radius is infinite. Termwise differentiation is what licenses substituting the series into (1); without it the recurrence would be a formal identity only.
The non-obvious move is in the Wronskian step. Abel's identity makes constant, so its value can be read off as , where only the leading terms matter. That is where enters: the limit is , and at the two series coincide and the argument gives no second solution.
Resonance at integer order
Theorem 2.5.2
(The negative-root obstruction for positive integers)
For an integer , no solution of (1) with has a Frobenius expansion with and a convergent power series near zero. The positive-root construction still gives a solution.
Proof
(The recurrence fails at a specific coefficient)
Take . Then (2) reads
At the coefficient is nonzero, so . For all denominators are nonzero; induction gives the odd coefficients zero and all even coefficients through nonzero when . In fact every odd coefficient vanishes, since its denominator is nonzero. At the recurrence instead requires , a contradiction. This rules out this Frobenius form; it does not by itself construct the second solution.
Common mistake
(Reading an integer root difference as an obstruction)
The distinction between an integer order and an integer difference of roots matters. For the roots differ by one, yet both series above exist: at , the equation is and leaves free; choosing gives the stated even series. An integer difference signals a possible obstruction; it is not by itself proof that one occurs.
Worked example 2.5.3
(Order three-halves: a gap of three with no obstruction)
For the roots differ by . At the recurrence (2) reads . At it gives , and at it reads , which holds because ; so is free and we choose . The even coefficients follow from :
so . These agree with the expansion of , since . Edge case: the resonance at is an odd index, where the recurrence is automatically consistent. The theorem gives .
Order zero and a logarithmic solution
At , the indicial root is repeated. The first solution is
By continuity we choose so that it is nonzero on the whole interval ; the integral below uses this entire zero-free interval.
Theorem 2.5.4
(A logarithmic second solution at order zero)
On the function
solves (1) with , satisfies , and so forms with a basis of solutions on . Moreover
Proof
(Reduction of order with the logarithm kept)
The integrand is at zero because , so the integral is well-defined. Its derivative gives
To verify the equation, write . The displayed derivative gives , so . Substituting into (1) and using the equation for proves that solves (1) for . Its Wronskian is nonzero, so the two solutions are independent on . Finally, the integral is and is bounded, which gives the asymptotic form.
The asymptotic form shows that the logarithm cannot be discarded.
Worked example 2.5.5
(The first terms of the logarithmic solution)
Write with . Then
since and . Hence , and the integral is . Multiplying by ,
because . The non-logarithmic part starts at , in agreement with the theorem.
Common mistake
(Carrying the quotient formula through a zero of )
A second solution can be continued from any positive point by the regular initial-value theorem, but the displayed quotient formula is used only where is nonzero. It must not be read through a later zero of without further work.
Worked example 2.5.6
(Half-integer order and elementary functions)
For , the two normalized solutions are
Expanding sine and cosine gives leading terms and with coefficient one, and substitution verifies the equation. Their Wronskian is , agreeing with . The integer difference between the roots has not forced a logarithm.
Counterexample mode
Two roots do not automatically give two series
For , the roots are and . The positive-root solution starts
For the negative root, (2) forces at and then forces at . This contradicts the required nonzero leading coefficient. Listing the two roots without checking the recurrence would invent a solution of a form that does not exist.
The results above do not prove the general Frobenius theorem or derive every integer-order Bessel solution. They give a complete convergence and independence argument for the specified noninteger solutions and a local second solution at order zero.
Exercises
Quick check
(Derive the coefficient of in and check the first three displayed terms against (1) when .)
Solution · Answer
The recurrence gives and . Substitution of cancels the constant, and coefficients; the remaining residual starts at order .
Quick check
(For , use (2) to derive the first three nonzero terms of and explain the obstruction for a leading series.)
Solution · Answer
With , and , giving . For , the equation gives and the equation requires , contrary to a leading term.
Quick check
(Verify directly that and solve (1) for , and compute their Wronskian.)
Solution · Answer
Substituting reduces (1) at to . Taking gives the two solutions. Their Wronskian is .
Quick check
(Why does imply independence on ? Does it mean the displayed integral formula for is valid across every positive zero of ?)
Solution · Answer
A constant linear dependence with nonzero coefficients would force their Wronskian to vanish identically; here it is nonzero throughout the interval. No: the displayed formula divides by and was defined only on a neighborhood with no zeros. Continuation of the solution through a positive point is a separate initial-value argument.
Quick check
(For , determine the Wronskian of the normalized solutions and explain why they form a basis.)
Solution · Answer
Since is positive and noninteger, the theorem gives . It is nonzero for every , so the solutions form a basis by linear ODE uniqueness.