AMS 212A Applied Mathematical Methods I Lecture 06 Copyright by Hongyun Wang, UCSC Recap of Lecture 5 Classification of boundary conditions Dirichlet Neumann Mixed Adjoint operator, self-adjoint operator Sturm-Liouville problem (in self-adjoint form) L (u ) = r ( x ) u , u ( a ) + u ( a ) = 0 u b + u b = 0 ( ) ( ) x ( a, b ) where the linear differential operator is L ( u ) ( p ( x ) u x )x + q ( x ) u The inner product is b u, v u ( x ) v ( x ) r ( x ) dx a Results of Sturm-Liouville theory: u, 1 1 L (v) = L (u ), v r ( x) r ( x) (that is, 1 L ( i ) is self-adjoint). r ( x) All eigenvalues are real. For each eigenvalue, we can find a real eigenfunction. Eigenfunctions for different eigenvalues are orthogonal to each other. Results of Sturm-Liouville theory (continued) Since eigenvalues and eigenfunctions are both real, we only need to work with real functions. All eigenvalues are simple. (i.e., one corresponding eigenfunction for each eigenvalue.) -1- AMS 212A Applied Mathematical Methods I Suppose u(x) and v(x) are eigenfunctions corresponding to eigenvalue . Claim: u(x) and v(x) are proportional to each other. Proof: L (u ) = r ( x ) u L (v) = r ( x ) v ==> u L ( v ) vL ( u ) = u( r ( x ) v ) v( r ( x ) u ) = 0 On the other hand, using the differential form of L(•), we write ( ) ( u L ( v ) vL ( u ) = u ( pvx )x + qv v ( pu x )x + qu ) = u( pvx )x v( pu x )x (Using Lemma 1 of Lecture 5) = ( p ( uvx vu x ))x Combining these two results, we have ( p (uv ==> x vu x ))x = 0 p ( uvx vu x ) = const Since both u(x) and v(x) satisfy the boundary conditions, we have ( u v vu ) =0 (Lemma 2 of Lecture 5) x=b ==> p ( u v vu ) = 0 , ==> u v vu = 0 , ==> v = 0 , u ==> v = const u x [ a, b ] x [ a, b ] x [ a, b ] End of proof Eigenvalue sequence Eigenvalues form an unbounded strictly increasing infinite sequence 0 < 1 < 2 < < n < -2- AMS 212A Applied Mathematical Methods I lim n = + n+ We will prove this after we describe the completeness of eigenfunctions. Completeness of eigenfunctions The set of eigenfunctions {v0 ( x ) , v1 ( x ) , v2 ( x ) , …, vn ( x ) , …} is a complete basis for all piecewise continuous functions on [a, b]. That is, any piecewise continuous function ƒ(x) can be expanded as f ( x ) = bn vn ( x ) n=0 where coefficient bn is given by b f , vn bn = = vn , vn f ( x ) v ( x ) r ( x ) dx n a b ( v ( x )) r ( x ) dx 2 n a Below we do 3 things: prove the eigenvalue sequence, introduce the Rayleigh quotient, and then prove the completeness Proof of eigenvalue sequence Eigenvalue and eigenfunction u(x) satisfy the Sturm-Liouville equation ( p ( x)u ) x x x ( a, b ) + q ( x )u = r ( x )u , and boundary conditions u ( a ) + u ( a ) = 0 u (b ) + u (b ) = 0 The key step of the proof is Prufer substitution: p ( x ) u x ( x ) = ( x ) cos ( ( x )) u ( x ) = ( x ) sin ( ( x )) Before the Prufer substitution, we have a second order linear ODE for u(x). After the Prufer substitution, (as we will see) we have -3- AMS 212A Applied Mathematical Methods I a first order non-linear ODE system for (x) and (x). Let us look at the advantages of Prufer substitution. Advantage #1: ( x ) 0 , x [ a, b ] Suppose ( x0 ) = 0 at some point x0 in [a, b]. ==> p ( x0 ) u x ( x 0 ) = 0 u ( x0 ) = 0 p(x0) 0 ==> u x ( x 0 ) = 0 u ( x0 ) = 0 That is, u(x) satisfies the zero initial conditions at x0. The Sturm-Liouville equation is a second order ODE of u(x) ==> u ( x) 0 (due to the zero initial conditions at x0) In the Sturm-Liouville problem, we are only interested in non-trivial solutions. Advantage #2: Boundary conditions affect only Boundary condition at x = a: u ( a ) + u ( a ) = 0 (a) cos ( ( a )) = 0 p (a) ==> ( a ) sin ( ( a )) + ==> p ( a ) sin ( ( a )) + cos ( ( a )) = 0 ==> sin ( ( a ) ) = 0 where satisfies cos ( ) = p (a) ( p ( a )) 2 + 2 , sin ( ) = ( p ( a )) 2 + 2 The boundary condition becomes (a) = n ==> (a) = + n We can shift the whole function (x) to write the boundary condition at x = a as -4- AMS 212A Applied Mathematical Methods I BC at x = a : ( a ) = , 0 < After the shifting, we write the boundary condition at x = b as BC at x = b : ( b ) = μ + n , 0< μ Note: we use shifting and pick suitable value of n to make 0 < and 0 < μ , which is important in the proof of Lemma 1 and proof of eigenvalue sequence. Advantage #3: The evolution equation of (x) is independent of (x). (x) satisfies the first order non-linear ODE x = ( r ( x ) + q ( x )) sin 2 + 1 cos 2 p( x) Key steps in derivation: 2 equations for (x) and (x): Equation 1: ( p ( x ) u ) + ( r ( x ) + q ( x )) u = 0 (the Sturm-Liouville equation) ==> ( cos ) + ( r ( x ) + q ( x )) sin = 0 x x x Equation 2: cos = p ( x ) u x = p ( x ) ( sin )x … (See Exercise #6 for derivation). Lemma 1: Let (x, ) be the solution of x = ( r ( x ) + q ( x )) sin 2 + 1 cos 2 p( x) (a) = , 0 < where p(x) > 0 and r(x) > 0 for x in [a, b] (from the Sturm-Liouville problem). Then, (x, ) has the 3 properties below. For x > a, a) (x, ) is a strictly increasing function of . b) c) lim ( x, ) = + + lim ( x, ) = 0 (For proof of Lemma 1, see Appendices of Lecture 6 in a separate PDF file). -5- AMS 212A Applied Mathematical Methods I Now we use the results of Lemma 1 to prove the eigenvalue sequence. lim ( b, ) = 0 < μ (since 0 < μ ) ==> μ 0 < ( b, ) < μ for negative and large. As increases to infinity, (b, ) also increases to infinity. At some value, 0, we will have (b, 0) = μ. ==> 0 is the smallest value of for which (b, 0) = μ + n is satisfies. ==> 0 is the smallest eigenvalue. As increases further, at some value, 1, we will have (b, 1) = μ + ; ==> 1 is the smallest eigenvalue after 0. At some value, 2, we will have (b, 2) = μ + 2; ==> 2 is the smallest eigenvalue after 1. At some value, 3, we will have (b, 3) = μ + 3; ==> 3 is the smallest eigenvalue after 2. … Thus, we have a strictly increasing sequence of eigenvalues 0 < 1 < 2 < < n < Next we prove lim n = + . n+ Since { n} increases monotonically, we have either lim n = + or lim n = finite . n+ Suppose lim n = ( c) = finite . We have n+ ( c ) > n ==> ==> for any n ( ) ( b, ( ) ) = + , b, ( c) > ( b, n ) = μ + n c for any n which contradicts with ( c) = finite . End of proof of eigenvalue sequence Rayleigh quotient Definition: -6- n+ AMS 212A Applied Mathematical Methods I Rayleigh quotient is defined as R (u ) = u, 1 L (u ) r ( x) u, u b = u L ( u ) dx a b u r ( x ) dx 2 a Rayleigh quotient on eigenfunctions: Suppose 0 < 1 < 2 < < n < is the eigenvalue sequence and {v ( x ), v ( x ), v ( x ), …, v ( x ), …} 0 1 2 n is the sequence of corresponding eigenfunction. It is straightforward to verify that Rayleigh quotient satisfies R ( vn ) = n , n = 0, 1, 2, … N R bn vn = n=0 N b 2 vn , vn n n n=0 N b 2 n vn , vn n=0 N 0 = min R bn vn {bn } n = 0 (See Exercise #6 for derivation). This last result motivates the Rayleigh principle below. Rayleigh principle: Let u ( a ) + u ( a ) = 0 2 C BC = u ( x ) u C 2 and satisfies u (b ) + u (b ) = 0 2 Consider the minimization of R(u) over C BC . Let u0 = arg min R ( u ) 2 uCBC , u0 Claim 1: u0 is an eigenfunction corresponding to eigenvalue 0 and 0 = min 2 uCBC , u0 R (u ) Proof of Claim 1: Consider a function of a scalar variable s -7- AMS 212A Applied Mathematical Methods I g ( s ) = R ( u0 + s u ) 2 where u C BC 2 By definition of u0, we have g ( 0 ) =0 for any u C BC The derivatives of numerator and denominator of R(u0 + s u) are d u0 + s u, u0 + s u ds = 2 u, u0 s=0 1 d u0 + s u, L ( u0 + s u ) r ( x) ds = 2 u, s=0 1 L ( u0 ) r ( x) (See Exercise #6 for derivation). The derivative of R(u0 + s u) has the expression 2 1 d R ( u0 + s u ) = L ( u0 ) + R ( u0 ) u, u0 u, u0 , u0 r ( x) ds s=0 = 2 u0 , u0 u, 1 L ( u0 ) + R ( u0 ) u0 r ( x) (See Exercise #6 for derivation). 2 The derivative of R(u0 + s u) is zero for any u C BC . d 2 R ( u0 + s u ) = 0 for any u C BC ds s=0 ==> u, 1 2 L ( u0 ) + R ( u0 ) u0 = 0 for any u C BC r ( x) ==> 1 L ( u0 ) + R ( u0 ) u0 = 0 r ( x) ==> L ( u0 ) = R ( u0 ) r ( x ) u0 ==> R(u0) is an eigenvalue and u0 is a corresponding eigenfunction. Next we show R ( u0 ) = 0 . By definition of u0, we have R ( u0 ) = min R ( u ) 2 uCBC ==> R ( u0 ) R ( v0 ) = 0 (v0 is an eigenfunction corresponding to 0). Since 0 is the smallest eigenvalue and R(u0) is an eigenvalue, we have -8- AMS 212A Applied Mathematical Methods I 0 R ( u 0 ) Combining these two results, we conclude 0 = R ( u0 ) = min R ( u ) 2 uCBC End of proof of Claim 1 2 Next we look at the minimum of Rayleigh quotient over subset of C BC . Let WN = span {v0 , …, vN } { 2 WN = u u C BC and u WN } Consider the minimization of R(u) over WN . Let u N +1 = arg min R ( u ) uWN , u0 Claim 2: uN+1 is an eigenfunction corresponding to eigenvalue N+1 and N +1 = min uWN , u0 R (u ) (For proof of Claim 2, see Appendices of Lecture 6 in a separate PDF file) Rayleigh principle provides a way of approximating the lowest few eigenvalues, especially the smallest eigenvalue. Example: u = u u ( 0 ) = 0 , u (1) = 0 This is a Sturm–Liouville problem with p ( x ) = 1 , r ( x ) = 1 , q ( x ) = 0 , L ( u ) = u We solved it previously. The eigenvalues are n = ( n + 1) 2 , 2 n = 0, 1, 2, … Now we use Rayleigh principle to approximate the smallest eigenvalue. (Skip the details in Lecture) We try w ( x ) = x (1 x ) (it has to satisfy the boundary conditions) -9- AMS 212A Applied Mathematical Methods I The numerator and denominator of Rayleigh quotient are 1 1 1 1 L ( w ) = w ( x ) w ( x ) dx = 2 x (1 x ) dx = r ( x) 3 0 0 w, 1 w, w = ( w ( x )) 1 2 0 dx = x 2 (1 x ) dx = 2 0 1 L (w) r ( x) w, ==> R (w) = ==> 0 = min R ( u ) R ( w ) = 10 w, w 1 30 1 3 = 10 1 30 = 2 uCBC The true value is 0 = 2 9.87 . Proof of the completeness of eigenfunctions For a piecewise continuous function ƒ(x), we show that N lim f bn vn = 0 N n=0 where f , vn vn , vn bn = b u, v = u ( x ) v ( x ) r ( x ) dx a u = u, u The proof consists of results 1-4 below. Result 1: N f bn vn v j n=0 for 0 j N This follows directly from the definition of bn. Result 2: N N f bn vn = min f cn vn n=0 {cn } n=0 - 10 - AMS 212A Applied Mathematical Methods I This follows directly from Result 1 above N N N N f b v v , f b v + n n n n n n + n vn n=0 n=0 n=0 n=0 N N = f bn vn , f bn vn + n=0 n=0 ==> N N n=0 n=0 N n vn , n=0 N v n n n=0 f cn vn f bn vn Result 3: We only need to show that for a piecewise continuous function ƒ(x), we have N for any > 0 there exists N and {cn} such that f cn vn < (***) n=0 Result 4: 2 is dense in the set of piecewise continuous functions, we only need to show Since C BC 2 that for f ( x ) C BC , we have N for any > 0 there exists N and {cn} such that f cn vn < n=0 For proof of (***), see Appendices of Lecture 6 in a separate PDF file. - 11 - (***)