Sturm-Liouville Theory: The Master Eigenvalue Problem Behind Special Functions

The precise statement: a regular Sturm-Liouville problem

A regular Sturm-Liouville problem on a finite interval [a,b] is the eigenvalue equation

−(p(x) u′(x))′ + q(x) u(x) = λ w(x) u(x),   a ≤ x ≤ b,

together with separated boundary conditions α₁u(a) + α₂u′(a) = 0 and β₁u(b) + β₂u′(b) = 0 (with (α₁,α₂) ≠ 0, (β₁,β₂) ≠ 0). The regularity hypotheses are essential: p ∈ C¹[a,b] with p(x) > 0, w ∈ C[a,b] with w(x) > 0, and q ∈ C[a,b] real. Then:

• The eigenvalues form an increasing real sequence λ₁ < λ₂ < λ₃ < ⋯ → +∞, each simple (one-dimensional eigenspace).
• The eigenfunctions {φₙ} are orthogonal in the weighted inner product ⟨f,g⟩ = ∫ₐᵇ f g̅ w dx, and after normalization form a complete orthonormal basis of L²([a,b], w dx).
• φₙ has exactly n−1 zeros in the open interval (a,b) — Sturm's oscillation theorem.

Completeness is the deep claim: every f ∈ L²(w) equals ∑ₙ ⟨f,φₙ⟩ φₙ in the L² norm.

The picture: self-adjointness forced by a perfect derivative

The whole theory rests on one integration-by-parts identity, Lagrange's identity. For the operator ℒu = −(p u′)′ + q u,

∫ₐᵇ (ℒu · v̅ − u · ℒ̅v) dx = −[ p (u′ v̅ − u v̅′) ]ₐᵇ,

and the boundary term on the right vanishes precisely because of the separated boundary conditions. So ⟨ℒu, v⟩ = ⟨u, ℒv⟩: the operator is symmetric (formally self-adjoint) on its domain. That single fact is the engine. Symmetry immediately forces eigenvalues to be real (⟨ℒφ,φ⟩ = λ‖φ‖² = ⟨φ,ℒφ⟩ = λ̅‖φ‖²) and eigenfunctions with different eigenvalues to be orthogonal ((λₘ − λₙ)⟨φₘ,φₙ⟩ = 0). Writing the ODE in the divergence form −(p u′)′ is not cosmetic — it is exactly what makes the boundary term a perfect bracket. Any second-order operator au″ + bu′ + cu can be converted to this self-adjoint form by multiplying through by the integrating factor (1/a)·exp(∫ b/a dx), which is why the theory is universal rather than special.

Key idea of the proof: invert the operator to get compactness

Symmetry gives orthogonality and real eigenvalues cheaply, but completeness — that the eigenfunctions span everything — is harder, because ℒ is an unbounded operator and unbounded symmetric operators need not have any complete eigenbasis. The trick is to invert. Shift λ so that 0 is not an eigenvalue; then ℒ has a Green's function G(x,ξ), built from two solutions y₁, y₂ satisfying the left/right boundary conditions, glued by the Wronskian jump condition. The inverse operator

(ℒ⁻¹ f)(x) = ∫ₐᵇ G(x,ξ) f(ξ) w(ξ) dξ = K f

is an integral operator with a continuous, symmetric kernel. By the Arzelà–Ascoli / Hilbert–Schmidt argument, K is a compact self-adjoint operator on L²(w). Now invoke the spectral theorem for compact self-adjoint operators: K has a complete orthonormal eigenbasis with eigenvalues μₙ → 0. Undoing the inversion, φₙ are eigenfunctions of ℒ with λₙ = 1/μₙ → ∞, and completeness of K's basis is completeness for ℒ. Compactness, obtained by inverting, is the whole game.

Canonical example: Fourier series as the simplest case

Take p ≡ 1, q ≡ 0, w ≡ 1 on [0, π] with Dirichlet conditions u(0) = u(π) = 0. The problem collapses to −u″ = λu, the classical vibrating-string equation. Solutions are u(x) = sin(√λ x); the boundary condition u(π)=0 forces √λ π = nπ, so

λₙ = n²,   φₙ(x) = √(2/π) · sin(n x),   n = 1, 2, 3, …

Everything the theorem promises is visible: eigenvalues n² are real, simple, increasing to +∞; φₙ has exactly n−1 interior zeros; the {sin nx} are orthogonal; and completeness is precisely the classical Fourier sine-series theorem. A less trivial case: Legendre's equation ((1−x²)u′)′ + λu = 0 on [−1,1] is Sturm-Liouville with p = 1−x², w = 1. Requiring boundedness at the singular endpoints x = ±1 selects λ = n(n+1) and the eigenfunctions are the Legendre polynomials Pₙ(x) — the same machine, run on a singular problem, produces an entire family of orthogonal polynomials.

Why the hypotheses matter — and where they break

Positivity of p and w is not optional. If p vanishes inside (a,b), the leading term degenerates, the ODE becomes singular, and eigenvalues can cease to be discrete or bounded below. Separated (or periodic) boundary conditions are what kill the Lagrange boundary term; drop them and ℒ is no longer symmetric, so eigenvalues may go complex and the eigenfunctions lose orthogonality. Simplicity of eigenvalues also depends on separated conditions: the periodic problem u(a)=u(b), u′(a)=u′(b) is self-adjoint but eigenvalues can be double — e.g. −u″ = λu on [0,2π] periodic gives λ = n² with the two-dimensional span {cos nx, sin nx}. On infinite intervals or with singular endpoints (Hermite on ℝ, Bessel at 0) the resolvent may fail to be compact and continuous spectrum appears; the correct framework is then Weyl–Titchmarsh theory and the limit-point/limit-circle dichotomy. The theory connects directly to the spectral theorem for self-adjoint operators in Hilbert space, of which it is the founding archetype.

Why it matters: the master template for eigenfunction expansions

Sturm-Liouville theory is the reason separation of variables works for the heat, wave, and Schrödinger equations. Separating a PDE on a domain produces an ODE eigenvalue problem in each coordinate; because that problem is Sturm-Liouville, its eigenfunctions form a complete basis, so the full PDE solution is a convergent series ∑ cₙ φₙ(x) Tₙ(t). This is exactly how you get Fourier series on an interval, Fourier–Bessel series in a cylinder, and Legendre/spherical-harmonic expansions on a sphere. In quantum mechanics the time-independent Schrödinger equation −ψ″ + V ψ = E ψ is a Sturm-Liouville problem: discreteness of the spectrum is the quantization of energy, and completeness is the statement that energy eigenstates span the Hilbert space. Historically the theory launched spectral analysis: it is the finite, tractable ancestor of the Hilbert-space spectral theorem, of Weyl's work on singular operators, and of the modern inverse spectral problem ("can you hear the shape of a drum?"). Nearly every eigenfunction expansion in applied mathematics is a Sturm-Liouville expansion in disguise.

Classical special functions realized as regular or singular Sturm-Liouville eigenproblems −(pu′)′ + qu = λ w u.

Special function	Interval [a,b]	p(x)	weight w(x)	eigenvalues λₙ
Fourier (sine/cosine)	[0, π]	1	1	n²
Legendre Pₙ	[−1, 1]	1 − x²	1	n(n+1)
Chebyshev Tₙ	[−1, 1]	√(1 − x²)	1/√(1 − x²)	n²
Hermite Hₙ	(−∞, ∞)	e^(−x²)	e^(−x²)	2n
Laguerre Lₙ	[0, ∞)	x e^(−x)	e^(−x)	n
Bessel Jᵥ (order ν)	[0, R]	x	x	j²ᵥ,ₙ / R²