Motivation
Initial conditions prescribe a function and its derivative at one point. Boundary conditions can instead prescribe values at two different points. The difference changes the question of solvability. Every initial pair determines a solution of , but a nonzero solution satisfying exists only for certain parameters. Those special parameters are eigenvalues, and their solutions become building blocks for more general functions.
Before considering boundary conditions, we need a way to construct solutions when elementary formulas are unavailable. A power series turns an equation into a recurrence for coefficients. Near a singular point, an ordinary power series may fail even when the equation still has a controlled structure; a Frobenius series introduces a leading power . Both methods require more than formal coefficient matching. One must identify the domain, justify convergence, and determine whether the constructed solutions are independent.
This note links local series construction to regular boundary eigenvalue problems. The regular theory below deliberately keeps finite endpoints and positive continuous weights; singular endpoint problems require additional hypotheses.
Definitions
Consider
An ordinary point is one at which and are analytic on a neighborhood. A real analytic function is locally represented by a convergent real power series. In the convergence statement below, it is convenient to assume analytic extensions to a complex disk; the real conclusions follow by restriction. The leading coefficient of a general equation must first be divided out where it is nonzero. Zeros of that coefficient are candidates for singular points, but cancellation may remove an apparent singularity.
Definition 2.1.1
(Regular singular point and indicial polynomial)
The point is regular singular if and extend analytically to . On the side , a Frobenius trial solution is , with . Its lowest power gives the indicial equation
The restriction makes a nonintegral real power single-valued. On , one can work with powers of using a separate derivation. Complex powers require a branch of the logarithm. A solution on one side of a singular point does not automatically extend through it, and an initial value theorem at an ordinary point cannot be applied at the singular point itself.
Concept lensAlgebraic
Indicial roots as characteristic roots after a logarithmic change of variable
When and are constants , the substitution , gives and , so becomes
Its characteristic polynomial is exactly the indicial polynomial. Distinct roots give ; a repeated root gives . In the repeated-root example below, produces and . For nonconstant analytic , the indicial polynomial still uses only , the coefficients of this constant model.
A regular Sturm–Liouville expression on , with , is
where is real and continuously differentiable, are real and continuous, and are strictly positive on the closed interval. The weighted inner product is . A nonzero function satisfying the equation and the chosen homogeneous boundary conditions is an eigenfunction; its parameter is an eigenvalue. In the main spectral statement we use Dirichlet conditions .
Separated real boundary conditions can more generally specify a nontrivial linear relation between and at each endpoint. Such conditions also make the Green boundary bracket vanish between admissible functions. Their energy formula may contain endpoint terms, however. The simpler Dirichlet assumptions are kept explicit whenever positivity is claimed below.
Theorem/Proposition
Theorem 2.1.2
(Analytic initial value solutions at an ordinary point)
Suppose and extend analytically to the complex disk . For every pair , , there is a unique analytic solution on that disk. Its Taylor coefficients are obtained by substituting a power series into the differential equation. The radius of the resulting series is at least ; a singularity of the coefficients on the boundary need not be a singularity of every individual solution.
Theorem 2.1.3
(Reality, orthogonality, and energy for Dirichlet eigenfunctions)
For the regular Sturm–Liouville problem with Dirichlet conditions, every eigenvalue is real. Eigenfunctions belonging to distinct eigenvalues are orthogonal in . Every eigenfunction satisfies
In particular, if , then every eigenvalue is strictly positive. Each eigenvalue has a one-dimensional eigenspace.
The last conclusion concerns geometric multiplicity. A nonzero solution vanishing at has a nonzero derivative there, because a zero value and zero derivative would force the zero solution. Thus its derivative fixes its scalar multiple. The theorem does not yet establish that eigenvalues exist, exhaust a sequence, or that eigenfunctions form a complete basis. These are further parts of regular Sturm–Liouville theory and are not proved in this note.
Proof sketch or proof idea
Proof
(Picard iteration for the analytic system)
For the analytic theorem, introduce and write with analytic matrix . On each closed disk strictly inside the coefficient disk, the integral iteration along paths from has successive differences bounded by a constant times . The factorial makes the series of iterates converge uniformly. The limit is analytic and satisfies the integral equation, hence the differential equation. Uniqueness follows either from the same estimate applied to a difference or from the coefficient recurrence. Constructing on every smaller disk yields the solution on the whole open disk. This is why formal substitution becomes a convergent method under analyticity, whereas a merely smooth coefficient need not give a convergent Taylor representation.
Proof
(The lowest power gives the indicial equation)
The indicial polynomial follows directly from the lowest power. With , the three lowest contributions to have coefficient at . Since is nonzero, the coefficient must vanish. Higher coefficients then impose a recurrence. This calculation is necessary but does not alone guarantee two Frobenius solutions. When the roots differ by an integer, an obstruction in the recurrence can force a logarithm. Repeated roots also require a separate second-solution construction.
Proof
(Energy identity, Green's identity and initial-value uniqueness)
For the spectral theorem, integrate by parts against . The Dirichlet values remove the endpoint term and give
The denominator is positive because and is continuous and nonzero. The right hand side is real, proving reality and the quotient. If and the numerator were zero, positivity of would force , so would be constant and the boundary conditions would force . This contradiction proves strict positivity.
For eigenfunctions with real eigenvalues , integration by parts gives Green's identity
Distinct eigenvalues therefore imply orthogonality. Finally, take two eigenfunctions at the same parameter and subtract a scalar multiple so that both value and derivative vanish at . The initial value uniqueness theorem forces their difference to vanish everywhere. Each step uses regularity at the endpoints; none can simply be transferred to an infinite or singular endpoint without checking a replacement boundary argument.
Proof X-ray
Which hypothesis earns which conclusion?
The positive weight makes the denominator nonzero; real coefficients make the energy numerator real. Vanishing endpoint terms then gives the comparison between two eigenfunctions. Strict positivity of an eigenvalue needs the extra sign condition and the Dirichlet conditions, which exclude a nonzero constant of zero energy. Reality, orthogonality and positivity are related conclusions with distinct logical inputs.
Worked examples
Worked example 2.1.4
(A recurrence with three interlaced coefficient chains)
For , substitute . The constant coefficient gives , while for ,
The independent initial coefficients are and . With one obtains
with one obtains
The coefficient functions are entire, so both series converge for every complex . Their initial value pairs are independent, making them a fundamental pair. The congruence class of an index modulo three organizes the recurrence; one coefficient chain vanishes because . Writing only the first two terms would obscure that structure.
Worked example 2.1.5
(A repeated indicial root and a logarithm)
Consider on . The indicial polynomial is , so . A second copy of is not independent. Put ; direct differentiation reduces the equation to , or . Hence , and
The Wronskian of and is , which is nonzero on the stated interval. Both functions tend to zero as , but that shared limit is not an initial condition at an ordinary point: the original equation is singular there. Even specifying a limiting value does not restore the ordinary-point uniqueness theorem.
Counterexample mode
Integer root difference does not always force a logarithm
For , the indicial roots are and . Their difference is the integer , yet the two powers and both solve the equation and are independent on . Thus the correct rule is that an integer difference may require a logarithmic second solution. The recurrence or another construction must decide. A rule that inserts a logarithm solely from the root difference would give a false answer here.
Worked example 2.1.6
(Dirichlet eigenvalues and weighted normalization)
Solve on with zero endpoint values. The energy theorem excludes . Write with . The left boundary condition leaves , and the right one requires . Thus
Since , the normalized functions are . Distinct functions are orthogonal either by a product-to-sum computation or by Green's identity. If a function is projected onto a finite selection of these modes, the coefficient of is . Computing these finite projection coefficients requires no assertion yet about convergence of an infinite expansion.
Common mistakes
Common mistake
(Formal series and counting indicial roots)
Formal series manipulations do not prove convergence. At an ordinary analytic point the theorem supplies convergence; at a regular singular point one needs the Frobenius convergence theorem with its hypotheses, or a separate argument for the particular recurrence. It is also unsafe to count two indicial roots as two independent solutions without checking their constructions. A repeated root gives only one leading power, while an integer difference leaves a case to investigate.
Common mistake
(A truncated series is not a solution)
A truncated series is an approximation, not an exact solution. Substituting the polynomial into the differential equation exposes a residual that measures its failure to solve the equation. The recurrence can cancel low powers of that residual while leaving higher ones. To turn the residual into a rigorous error bound, one also needs control of the tail or an integral estimate for the associated initial value problem. A small residual at one point does not bound error throughout a large interval. This matters especially near a singular endpoint, where coefficient growth can amplify an apparently small neglected term. Series truncation should therefore be accompanied by its interval and the basis of any stated accuracy claim.
Common mistake
(Weight, boundary conditions and normalization)
The weight in Sturm–Liouville orthogonality belongs in both the inner product and its normalization. For , ordinary unweighted orthogonality is generally the wrong statement. Positivity likewise depends on the boundary conditions and the sign of : Neumann conditions for admit the constant eigenfunction at eigenvalue zero, so the strict Dirichlet conclusion cannot be reused unchanged. Finally, eigenfunctions may be multiplied by any nonzero scalar; a boundary determinant identifies eigenvalues, not a preferred normalization. In expansions, distinguish pointwise equality of functions from equality in a weighted square-integrable space.
Summary
Analytic coefficients turn a Taylor recurrence into a convergent local solution construction. Regular singular equations require a leading power and an indicial equation, with logarithms decided by a further construction. Regular Dirichlet Sturm–Liouville problems supply a real energy quotient and weighted orthogonality. These conclusions use explicit endpoint and positivity hypotheses; they do not by themselves prove spectral completeness.
Exercises
Quick check
(Derive the power series solutions of from the recurrence, using the initial pairs and .)
Do not identify the functions until after finding the coefficients.
Solution · Answer
The recurrence is . The first initial pair gives , ; the second gives , . Thus the solutions are and , respectively. The factorial denominators give convergence everywhere, and their initial pairs prove independence.
Quick check
(Solve on and verify that your two solutions are independent.)
Its indicial polynomial has a repeated root.
Solution · Answer
The indicial polynomial is . Set ; substitution reduces to . Hence and . Their Wronskian is , which is nonzero for .
Quick check
(Find the Dirichlet eigenvalues of on , where .)
Compare with the constant-coefficient sine problem.
Solution · Answer
The quotient gives . Put . The endpoint conditions force for positive integers . Therefore , with eigenfunctions . The strict inequality excludes the constant and hyperbolic cases.
Quick check
(For a regular Dirichlet problem with , why can a nonzero eigenfunction not have eigenvalue zero? What changes for Neumann conditions when ?)
Use the integrated energy identity.
Solution · Answer
If , the energy identity and force . A constant satisfying the two Dirichlet values must vanish, contradicting the definition of eigenfunction. With Neumann conditions and , a nonzero constant satisfies both derivative conditions and . Zero is then an eigenvalue. The boundary conditions change the conclusion even though the differential expression is the same.