Chapter 24 Sturm-Liouville problem Speaker: Lung-Sheng Chien Reference: [1] Veerle Ledoux, Study of Special Algorithms for solving Sturm-Liouville and Schrodinger Equations. [2] 王信華教授, chapter 8, lecture note of Ordinary Differential equation Existence and uniqueness [1] Sturm-Liouville equation: d dy p x q x y Ew x y on interval 0, a dx dx Assumptions: 1 p x is continuous differentiable on closed interval 0, a , or say p C1 0, a , and p 0 on 0, a lim p x p 0 x 0 lim p x p a lim p x p 0 x 0 , and lim p x p a xa , and min p x : 0 x a pmin 0 xa 2 q x is continuous on closed interval 0, a , or say q C 0, a 3 w x C 0, a , and w 0 on 0, a d dy p x q x y Ew x y dx dx dy y 0 y0 , 0 y01 dx Proposition 1.1: Sturm-Liouville initial value problem under three assumptions. is unique on 0, a proof: as before, we re-formulate it as integral equation and apply “contraction mapping principle” x dy 1 p 0 y01 q s Ew s y s ds x 0 dx p x y x y0 p 0 y01 x 0 Let x 1 1 t ds q s Ew s y s ds dt 0 p t 0 p s C 0, a f : 0, a R is continuous be continuous space equipped with norm f max f x : 0 x a Existence and uniqueness [2] 1 C 0, a is complete under norm 2 define a mapping T : C 0, a C 0, a f max f x : 0 x a Ty x y0 p 0 y01 0 x by 1 t q s Ew s s s ds dt 0 p t T x T x 0 x extract supnorm T x T x 0 x 1 t q pmin 0 q T x T x T T q E w 2 pmin q E w 2 pmin T is a contraction mapping if x 1 1 t ds q s Ew s y s ds dt 0 0 p s p t E w 2 pmin x 2 a 2 E w ds dt a2 1 Existence and uniqueness Fundamental matrix Sturm-Liouville equation: d dy p x q x y Ew x y on interval 0, a dx dx Transform to ODE system by setting 0 d y dx z q x Ew x dy dx 1 y1 0 1 y2 0 0 y1 y2 y , p x has two fundamental solutions , satisfying z z1 0 0 z2 0 1 z z 1 2 0 d dy p x q x y Ew x y dx dx is dy y 0 y0 , 0 y01 dx Solution of initial value problem Definition: fundamental matrix of satisfying z x p x d dy y p x q x y Ew x y is x 1 dx dx z1 y 0 0 1 z1 0 y2 0 1 z2 0 1 Question: Can we expect that we have two linear independent fundamental solutions, say y1 x y2 x 0 on 0, a y x y0 y1 x p 0 y01 y2 x 0 y2 y1 z2 p x y1 p x y2 y2 Abel’s formula Consider general ODE system of dimension two: y d det x det 1 dx z1 [1] d y a11 x a12 x y dx z a21 x a22 x z y2 y1 det z2 z1 y2 z2 y1 z1 with x y2 , 0 I2 z2 ( from product rule) y det 1 z1 y2 a11 y1 a12 z1 a11 y2 a12 z2 a11 y1 a11 y2 y1 det det a11 det z2 z1 z2 z2 z1 z1 y det 1 z1 y2 y1 y2 y2 y det det 1 a22 det z2 a y a z a y a z a z a z 21 1 22 1 21 2 22 2 22 1 22 2 y2 a det z2 11 d det x a11 x a22 x det x trA det x dx First order ODE det x det 0 exp trA s ds x 0 dy x dx y dy x y dx x dy y0 y 0 s ds y log det 0 0 det x 0 implies y1 x , y2 x are linearly independent x y x s ds y 0 0 Abel’s formula y d y a11 x a12 x y a x a x A x dx z 21 22 z z y1 z1 with fundamental matrix x d dy p x q x y Ew x y dx dx z x p x y2 , det 0 0 z2 Abel’s formula: det x det 0 exp [2] trA s ds x 0 0 d y dx z q x Ew x dy dx 1 y p x z 0 det x det 0 x 0, a Definition: Wronskian W 1 , 2 y x 1 z1 since trA 0 det 1 2 , then fundamental matrix of Sturm-Liouville equation can be expressed as 1 2 y2 y1 z2 p x y1 p x W y1 , y2 p x y2 y2 In our time-independent Schrodinger equation, we focus on eigenvalue problem d dy p x q x y Ew x y dx dx Definition: boundary condition 1 p x , q x V , w x 1 2 1 d2 V x x E x , 0 0 2 2 dx 0 0 is called Dirichlet boundary condition Question: What is “solvability condition” of Sturm-Liouville Dirichlet eigenvalue problem? Dirichlet eigenvalue problem Dirichlet eigenvalue problem: [1] d dy p x q x y Ew x y , y 0 y a 0 dx dx Observation: 1 If , E is eigen-pair of Dirichlet eigenvalue problem, then C 0, a ,in fact, C 2 0, a dp d dx is continuous on closed interval 0, a q Ew d dx 2 dx 2 Under assumption | : analyitc p, q, w : analyitc Exercise: 2 p W a 0 w x C 0, a , and w 0 on 0, a , we can define inner-product x x w x dx where x is complex conjugate of 1 If w is not positive, then such definition is not an “inner-product” 2 | x is Dirac Notation, different from conventional form used by Mathematician b Thesis of Veerle Ledoux: yi , y j a yi y j wdx Matrix computation: u H v u1 u 2 v1 u3 v2 v 3 u v Dirac Notation: | Dirichlet eigenvalue problem 3 4 1 d d p x q x , then it is linear and w x dx dx Define differential operator L d dy p x q x y Ew x y , y 0 y a 0 dx dx | L Green’s identity: | L | L | L W W W a p p Ly Ey, y 0 y a 0 pW , |xx0a d d L wdx p x 0 0 dx dx a W | L W [2] d a |0 q x dx p dx | L W 0 d d p dx dx q x dx d a a d d |0 p q x dx 0 dx dx dx d a |0 dx a 0 d d p dx dx q x dx The same a hence a | L W d d p pW , |xx 0a dx 0 dx Dirichlet boundary condition: | L W | L W 0 0 a 0 0 a 0 Dirichlet eigenvalue problem 2 5 Definition: operator L is called self-adjoint on inner-product space L If Green’s identity is zero, say | L W | L W Matrix computation (finite dimension): [3] 0, a , w f : 0, a C : a 0 f wdx 2 0 Functional analysis (infinite dimension): ei Ae j Aij ei e j Aei A ji e j Matrix A is Hermitian (self-adjoint): Aij Aji | L | L W Lij W Lji operator L is self-adjoint: | L W | L W 0 Matrix A is Hermitian, then 1 A is diagonalizable and has real eigenvalue 2 eigenvectors are orthogonal AV V diag 1 , A V V , n V H V I : orthonormal operator L is Hermitian, then H ? 1 L is diagonalizable and has real eigenvalue 2 eigenvectors are orthogonal Dirichlet eigenvalue problem [4] d dy Suppose , E1 and , E2 are two eigen-pair of p x q x y Ew x y , y 0 y a 0 dx dx , E1 | L W L E1 and , E2 are two eigen-pair L E2 L E1 | L W E1 | W L E2 | L W E2 | W | L W E2 E1 | | L W | L W 0 W 0 E2 E1 | W Theorem 1: all eigenvalue of Sturm-Liouville Dirichlet problem are real <proof> L 1 d d p x q x w x dx dx L L E1 p, q, w : real E 1 L L L L L E1 Hence , E1 is eigen-pair if and only if , E1 is eigen-pair 0 E2 E 1 | E2 E1 , W 0 E1 E1 | 0 W E1 E1 E1 R Dirichlet eigenvalue problem [5] d dy Theorem 2: if , E1 and , E2 are two eigen-pair of p x q x y Ew x y , y 0 y a 0 dx dx If eigenvalue E1 E2 , then | W 0 : , are orthogonal in L2 0, a , w <proof> from Theorem 1, we know E1 and E2 are real 0 E2 E1 | E1 , E2 : real W 0 E2 E1 | E1 E2 W 0 | W Theorem 3 (unique eigen-function): eigenfunction of Sturm-Liouville Dirichlet problem is unique, in other words, eignevalue is simple. <proof> Abel’s formula d dy p x q x y Ew x y dx dx z x p x 0 d y dx z q x Ew x Let x p dy dx 1 y p x z 0 det x det 0 x 0, a L E suppose L E since trA 0 det 0 0 and 0 a 0 0 a 0 , then p det x 0 x 0, a c for some constant c Dirichlet eigenvalue problem [6] Theorem 4 (eigen-function is real): eigenfunction of Sturm-Liouville Dirichlet problem can be chosen as real function. <proof> suppose u 1v is eigen-function with eigen-value E , satisfying L E u 1v and 0 a 0 Hence u, E , v, E u 1v Lu Eu and Lv Ev L L u 0 u a 0 v 0 v a 0 are both eigen-pair, from uniqueness of eigen-function, we have v cu v cu 1 c 1 u we can choose real function u as eigenfunction So far we have shown that Sturm-Liouville Dirichlet problem has following properties 1 Eigenvalues are real and simple, ordered as E0 E1 E2 2 2 Eigen-functions are orthogonal in L 0, a , w with inner-product | 3 Eigen-functions are real and twice differentiable Exercise (failure of uniqueness): consider d2y Ey dx 2 y y , W a 0 x x w x dx , find eigen-pair and d d y y dx dx show “eigenvalue is not simple”, can you explain this? (compare it with Theorem 3) Dirichlet eigenvalue problem [7] Theorem 5 (Sturm’s Comparison theorem): let 1 , E1 , 2 , E2 be eigen-pair of Sturm-Liouville Dirichlet problem. suppose E2 E1 , then 2 is more oscillatory than 1 . Precisely speaking 1 2 Between any consecutive two zeros of 1 , there is at least one zero of 2 <proof> x1 1 2 Let x1 , x2 are consecutive zeros of 1 and 1 0 on x1 , x2 as left figure suppose 2 0 on x1 , x2 x2 1, E1 , 2 , E2 d1 d p x q x 1 Ew x 1 , 1 0 1 a 0 dx dx d2 d p x q x 2 Ew x 2 , 2 0 2 a 0 dx dx are eigen-pair, then define det x det 2 p2 1 pW 2 , 1 p1 1 x1 0 det x1 2 x1 p1 x1 1 x1 p2 x1 2 x1 p1 x1 0 1 x2 0 det x2 2 x2 p1 x2 1 x2 p2 x2 2 x2 p1 x2 0 IVT det c 0, c x1 , x2 Dirichlet eigenvalue problem d det x det 2 dx p2 2 1 det p p1 2 2 d det x det dx q E2 w 2 det x1 0 and 1 2 det p p1 2 [8] 1 1 p 1 d1 d p x q x 1 Ew x 1 dx dx d2 d p x q x 2 Ew x 2 dx dx x1 2 x2 1 E2 E1 w12 0 on x1 , x2 q E1w 1 d det x 0 on x1 , x2 implies det x 0 on x1 , x2 dx det c 0, c x1 , x2 Question: why 2 is more oscillatory than 1 , any physical interpretation? Sturm-Liouville equation Time-independent Schrodinger equation V x x E x 2 me d dy p x q x y E w x y dx dx Definition: average quantity of operator Ô over interval I x1 , x2 by Oˆ I ˆ yOywdx / y 2 wdx I I Dirichlet eigenvalue problem 1 d d p x q x T V w x dx dx Express operator L as L | L L E where | L d x2 p |x dx 1 I I x2 x1 E | [9] where T : kinetic operator, V : potential operator I d p dx dx 2 x2 x1 q 2 dx neglect boundary term L I E T I V I E where T I d p dx dx 2 x2 x1 x2 x1 and wdx 2 V I x2 x1 x2 x1 q 2 dx 2 wdx Heuristic argument for Theorem 5 1, E1 , 2 , E2 T T I 1 E1 2 E2 I V V Average value of I E2 E1 T I 2 T I d 2 d1 Average value of dx dx 2 is more oscillatory than 1 I 2 x2 x1 x2 d p 2 dx x1 dx 2 d p 1 dx dx 2 Prufer method [1] Sturm-Liouville Dirichlet eigenvalue problem: d dy p x q x y Ew x y , y 0 y a 0 dx dx Idea: introduce polar coordinate , in the phase plane y sin z py cos dy dx sin dz cos dx d cos dx sin d dx 0 d y dx z q x Ew x 1 y p x z 0 Objective: find eigenvalue E such that y, z py 2 y2 z2 tan y z d dx sin cos d dx dy cos dx sin dz dx d 1 cos 2 Ew q sin 2 dx p 1 d 1 Ew q sin cos dx p Exercise: check it Objective: find eigenvalue E such that y 0 y a 0 tan 0 tan a 0 d 1 cos 2 Ew q sin 2 dx p when solution x; E is found with condition tan 0; E tan a; E 0 Advantage of Prufer method: we only need to solve x then x 0 exp 1 Ew t q t sin t ; E cos t; E dt 0 p t Prufer method [2] Observation: 1 2 y2 z2 0 Suppose x0 0, 0 x0 a y x0 0, z x0 0 0 d y dx z q x Ew x 1 y p x z 0 has unique solution y x 0, z x 0 with initial condition y x0 0, z x0 0 2 fixed x 0, x; E is a strictly increasing function of variable E d dy p x q x y Ew x y dx dx d2y 2 Ey dx y 0 0 p 1, q 0, w 1 d2y 2 Ey is Hook’s law dx has general solution y sin kx, k E , E 0 d 1 cos 2 Ew q sin 2 dx p d x; k1 cos 2 k12 sin 2 dx d x; k2 cos 2 k22 sin 2 dx p 1, q 0, w 1 k1 k2 d cos 2 E sin 2 dx and tan x; E slope : cos 2 k12 sin 2 cos 2 k22 sin 2 implies x; k1 x; k2 0 1 tan kx k Prufer method [3] Question: Given energy E, how can we solve d cos 2 E sin 2 , 0 0 numerically dx Forward Euler method: consider first order ODE dy f x , y , y 0 y0 dx 1 Uniformly partition domain 0, a as 0 x0 x1 x2 2 k Let y y xk and approximate continuous equation: x j jh 2 dy xk by one-side finite difference dy x y x h y x h d y2 c dx dx h 2 dx dy f x , y , y 0 y0 dx y x h y x dy x dx h k 1 yk discrete equation: y f xk , y k , y 0 y0 h xn a Exercise : use forward Euler method to solve angle equation d x; E cos 2 x; E E sin 2 x; E , 0; E 0 dx for different E = 1, 4, 9, 16 as right figure It is clear that x; E is a strictly increasing function of variable E Prufer method [4] y sin z py cos 3 d dy p x q x y Ew x y dx dx y 0 y a 0 y x 0 Although y sin 0 d 1 cos 2 Ew q sin 2 dx p 0 0 2 y2 z2 tan y z d 1 cos 2 Ew q sin 2 dx p tan 0 tan a 0 sin x; E 0 x; E n has a unique solution x; E . But we want to find energy E d 1 2 2 such that dx p cos Ew q sin , that is a; E n is constraint. 0 0, tan a 0 Example: for model problem 1.817 n d cos 2 E sin 2 , 0 0, E 3 dx Prufer method [5] 4 d 1 cos 2 Ew q sin 2 dx p y xi 0 xi n y sin y x 0 d 1 0 xi dx p xi x; E n 0 never decrease in a point where xi n number of zeros of y in 0, a number of multiples of in 0 , a n 1 n 1 n n x, E0 tan x /2 x y x; E0 /2 y x py x y x 0 tan x 0 y x 0 tan x x n Ground state y x; E0 has no zeros except end points. x 1 x n 2 Prufer method [6] x, E1 2 First excited state y x; E1 has one zero in 0, 3 / 2 d 1 cos 2 Ew q sin 2 dx p /2 x y x; E1 z py cos 0 d x j Ew q dx x model problem: x, E2 1 x j n 2 d xj E 0 dx second excited state y x; E1 has two zeros in 0, 3 5 / 2 2 3 / 2 conjecture: k-th excited state y x; Ek has k zeros in /2 x y x; E2 x 0, Prufer method [7] Theorem 6 : consider Prufer equations of regular Sturm-Liouville Dirichlet problem d dy p x q x y Ew x y , y 0 y a 0 dx dx Let boundary values satisfy following normalization: p C1 0, a , q, w C 0, a and p 0, w 0 d 1 cos 2 Ew q sin 2 dx p 0 0, a Then the kth eigenvalue Ek satisfies 0, Ek 0, a, Ek k Moreover the kth eigen-function y x, Ek 0 has k zeros in 0, a Remark: for detailed description, see Theorem 2.1 in Veerle Ledoux, Study of Special Algorithms for solving Sturm-Liouville and Schrodinger Equations. Prufer method [8] Scaled Prufer transformation: a generalization of simple Prufer method y 1 sin S z py S cos d dx 1 2 y where S x; E 0 : scaling function, continuous differentiable 1 sin S d y S dx z 2 S z py S cos sin S A S cos S sin 1 A cos S cos S S sin d S 1 d y A dx dx z 2 S 1 sin sin S S cos S cos 1 S sin cos cos S sin S cos d S dx S sin Theorem 6 holds for scaled Prufer equations dy z dz , q x Ew x z dx p x dx d S Ew q 2 S cos 2 sin sin cos dx p S S 2 d S Ew q S sin 2 cos 2 dx p S S Exercise: use Symbolic toolbox in MATLAB to check scaled Prufer transformation. Prufer method [9] Scaled Prufer transformation Simple Prufer transformation d S Ew q 2 S cos 2 sin sin cos dx p S S d 1 cos 2 Ew q sin 2 dx p S 1 1 d 1 Ew q sin cos dx p 2 d S Ew q S sin 2 cos 2 dx p S S Recall time-independent Schrodinger’s equation 1.0542 1034 J s 2 d2 x V x E x 2m dx 2 me mass of electron 9.11031 kg x ay, or say x a dimensionless form E E, or say E V E 1 ma 2 ma 2 y 2 V ay ay 2 E ay 2 2 ma 2 reduce to 1D Dirichlet problem 1 d2 V x x E x 2 2 dx 0 0 p 1 ,q V,w 1 2 d dy p x q x y Ew x y , y 0 y a 0 dx dx Exercise 1 [1] Consider 1D Schrodinger equation with Dirichlet boundary condition 1 d2 V x x E x 2 2 dx 0 0 1 V x 1 2 x 0.1 V : repulsive potential 2 use standard 3-poiint centered difference method to find eigenvalue 2 1 1 1 2 2 2h 1 1 diag V E 2 0 number of grids = 50, (n = 50) 1 x1 x2 x3 xn x xn 1 0 E0 0.7597 E1 all eigenvalues ae positive, can you explain this? 2 Ground state x; E0 0 has no zeros in 0, 1st excited state x; E1 1 has 1 zero in 0, 2nd excited state x; E2 2 has 2 zero in 0, Check if k-th eigen-function has k zeros in 0, Exercise 1 2 [2] solve simple Prufer equation for first 10 eigenvalues 1 d2 V x x E x , 0 0 2 2 dx d 1 cos 2 Ew q sin 2 dx p 1 p ,q V,w 1 2 0; E 0 d 2 cos 2 E V sin 2 dx 0; E 0 E 0 : 9 Forward Euler method: N = 200 dy f x , y , y 0 y0 dx y k 1 y k f xk , y k , y 0 y0 h 1 , Ek k is consistent with that in theorem 6 2 , Ek increases gradually “staircase” shape with “plateaus” at , Ek j and steep slope around 1 , Ek j 2 Question: what is disadvantage of this “staircase” shape? Exercise 1 Forward Euler method: N = 100 [3] Forward Euler method: N = 200 , E9 12 under Forward Euler method with number of grids, N = 100, this is wrong! Question: Can you explain why x, E9 is not good? hint: see page 21 in reference Veerle Ledoux, Study of Special Algorithms for solving Sturm-Liouville and Schrodinger Equations. Exercise 2 Scaled Prufer transformation [1] d 1 S Ew q S Ew q S cos 2 sin 2 dx 2 p S p S S A x B x cos 2 C x sin 2 d S Ew q 2 S cos 2 sin sin cos dx p S S 1 d2 V x 1D Schrodinger: x E x , 0 0 2 2 dx d A x B x cos 2 C x sin 2 dx 0; E 0 p 1 ,q V,w 1 2 d 1 E V E V S 2 S 2S cos 2 sin 2 dx 2 S S S 0; E 0 if E V x 1 1 S x f E V x Suppose we choose E V x if E V x 1 1 if x 1 x if x 1 where f x dS Question: function f is continuous but not differentiable at x = 1. How can we obtain dx Although we have known V x dS 1 dV x dx 2S dx 1 2 x 0.1 dS x 0 on open set I1 x : E V x 1 and dx 2 on open set I 2 x : E V x 1 E 1 x x0 I1 I2 V x0 1 2 x0 0.1 2 E 1 x0 1 2 E 1 0.1 is continuous f x x Exercise 2 [2] solve simple Prufer equation for first 10 eigenvalues 1 d2 V x x E x , 0 0 2 2 dx d S Ew q 2 S cos 2 sin sin cos dx p S S p 1 ,q V,w 1 2 0; E 0 0.5 S x Choose scaling function 0.5 E V x E 0 : 9 d E V S 2 S cos 2 sin 2 sin cos dx S S 0; E 0 if E V x 1 if E V x 1 Forward Euler method: N = 100 Forward Euler method: N = 100 scaling function S NO scaling function S Compare both figures and interpret why “staircase” disappear when using scaling function