TheoremComplete

Sturm's Comparison and Oscillation Theorems

Sturm's theorems provide qualitative information about the zeros and oscillation of solutions to second-order linear ODEs without solving them explicitly.


Sturm's Comparison Theorem

Theorem9.5Sturm's comparison theorem

Consider two equations on an interval [a,b][a, b]:

y+q1(x)y=0andz+q2(x)z=0,y'' + q_1(x)y = 0 \quad \text{and} \quad z'' + q_2(x)z = 0,

where q1,q2q_1, q_2 are continuous and q1(x)q2(x)q_1(x) \leq q_2(x) on [a,b][a, b] (with strict inequality on a subset of positive measure). Let yy be a nontrivial solution of the first equation with consecutive zeros at α,β[a,b]\alpha, \beta \in [a,b].

Then every nontrivial solution zz of the second equation has at least one zero in (α,β)(\alpha, \beta).


Proof

Proof

Suppose for contradiction that z(x)>0z(x) > 0 on (α,β)(\alpha, \beta) (without loss of generality, adjusting sign if necessary). Since yy has consecutive zeros at α\alpha and β\beta, we may assume y(x)>0y(x) > 0 on (α,β)(\alpha, \beta).

Consider the Wronskian-like expression:

W(x)=y(x)z(x)y(x)z(x).W(x) = y(x)z'(x) - y'(x)z(x).

Differentiating: W(x)=yzyz=y(q2z)(q1y)z=(q1q2)yzW'(x) = yz'' - y''z = y(-q_2 z) - (-q_1 y)z = (q_1 - q_2)yz.

Since q1q2q_1 \leq q_2 and y,z>0y, z > 0 on (α,β)(\alpha, \beta): W(x)0W'(x) \leq 0 on (α,β)(\alpha, \beta), with strict inequality on a set of positive measure.

Evaluate WW at the endpoints:

  • At x=αx = \alpha: W(α)=y(α)z(α)y(α)z(α)=y(α)z(α)W(\alpha) = y(\alpha)z'(\alpha) - y'(\alpha)z(\alpha) = -y'(\alpha)z(\alpha). Since yy increases from zero at α\alpha, y(α)>0y'(\alpha) > 0, and z(α)>0z(\alpha) > 0 (or z(α)=0z(\alpha) = 0). If z(α)>0z(\alpha) > 0, then W(α)<0W(\alpha) < 0... but we need to be more careful.

Actually, W(α)=y(α)z(α)0W(\alpha) = -y'(\alpha)z(\alpha) \leq 0 (since y(α)0y'(\alpha) \geq 0 and z(α)0z(\alpha) \geq 0). Similarly, W(β)=y(β)z(β)W(\beta) = -y'(\beta)z(\beta). Since yy decreases to zero at β\beta, y(β)0y'(\beta) \leq 0 (actually y(β)<0y'(\beta) < 0 since y>0y > 0 between), so W(β)=y(β)z(β)0W(\beta) = -y'(\beta)z(\beta) \geq 0.

Thus W(β)0W(α)W(\beta) \geq 0 \geq W(\alpha). But W0W' \leq 0 on (α,β)(\alpha, \beta) implies WW is non-increasing, so W(β)W(α)W(\beta) \leq W(\alpha). Combining: WW is constant, so W=0W' = 0 a.e., hence (q1q2)yz=0(q_1 - q_2)yz = 0 a.e. Since y,z>0y, z > 0 on (α,β)(\alpha, \beta), this requires q1=q2q_1 = q_2 a.e., contradicting the hypothesis. \blacksquare


Sturm's Oscillation Theorem

Theorem9.6Sturm's separation theorem

Let y1y_1 and y2y_2 be two linearly independent solutions of y+q(x)y=0y'' + q(x)y = 0. Then between any two consecutive zeros of y1y_1, there is exactly one zero of y2y_2 (and vice versa).

In other words, the zeros of y1y_1 and y2y_2 interlace.

ExampleZero interlacing

For y+y=0y'' + y = 0: y1=cosxy_1 = \cos x has zeros at π/2+nπ\pi/2 + n\pi, and y2=sinxy_2 = \sin x has zeros at nπn\pi. The zeros alternate: 0,π/2,π,3π/2,2π,0, \pi/2, \pi, 3\pi/2, 2\pi, \ldots, confirming Sturm separation.

For Bessel functions J0(x)J_0(x) and Y0(x)Y_0(x) (solutions of the Bessel equation of order 00), their positive zeros also interlace. This is used in practice to locate zeros of Bessel functions.

RemarkApplications of Sturm's theorems
  1. Bounding zeros: Comparing y+q(x)y=0y'' + q(x)y = 0 with y+λy=0y'' + \lambda y = 0 (constant coefficient), Sturm's comparison theorem bounds the distance between zeros of solutions.

  2. Oscillation criteria: If q(x)+q(x) \to +\infty as xx \to \infty, then all solutions oscillate (have infinitely many zeros). If q(x)0q(x) \leq 0 for large xx, solutions are eventually monotone (non-oscillatory).

  3. Eigenvalue counting: In Sturm-Liouville theory, the nn-th eigenfunction has exactly n1n-1 zeros in the interval, which follows from Sturm oscillation theory.