Orthogonal Polynomials and Related Approximation Results

Orthogonal Polynomials and Related Approximation Results

Chapter 3 Orthogonal Polynomials and Related Approximation Results The Fourier spectral method is only appropriate for problems with periodic boundary conditions. If a Fourier method is applied to a non-periodic problem, it inevitably induces the so-called Gibbs phenomenon, and reduces the global 1 convergence rate toO(N − ) (cf. Gottlieb and Ors ag (1!""##. $onse%uently, one should not apply a Fourier method to problems with non-periodic boundary conditions. Instead, one should use orthogonal polynomials which are eigenfunc- tions of some singular &turm-'iouville problems. The commonly used orthogonal polynomials include the 'egendre, $hebyshev, (ermite and 'aguerre polynomials. The aim of this chapter is to present essential properties and fundamental appro)imation results related to orthogonal polynomials. These results serve as preparations for polynomial-based spectral methods in the forthcoming chapters. This chapter is organi ed as follows. In the *rst section, we present relevant prop- erties of general orthogonal polynomials, and set up a general framewor+ for the study of orthogonal polynomials. ,e then study the -acobi polynomials in &ect. ../, 'egendre polynomials in &ect. ... and $hebyshev polynomials in &ect. ..0. In Sect. ..1, we present some general appro)imation results related to these families of orthogonal polynomials. ,e refer to & ego2 (1!"1#, 3avis and 4abinowit (1!50# and Gautschi (/660# for other aspects of orthogonal polynomials. 3.1 Orthogonal Polynomials Orthogonal polynomials play the most important role in spectral methods, so it is necessary to have a thorough study of their relevant properties. Our starting point is the generation of orthogonal polynomials by a three-term recurrence relation, which leads to some very useful formulas such as the $hristoffel-3arbou) formula. ,e then review some results on eros of orthogonal polynomials, and present ef*cient algorithms for their computations. ,e also devote several sections to discussing some important topics such as Gauss-type %uadrature formulas, polynomial inter- polations, discrete transforms, and spectral differentiation techni%ues. -. &hen et al., Spectral Methods: Algorithms, Analysis and Applications, 0" &pringer &eries in $omputational 7athematics 01, 3OI 16.166"8!"5-.-106-"1601-" ., �c &pringer-9erlag :erlin (eidelberg /611 05 . Orthogonal ;olynomials and 4elated <ppro)imation 4esults 3.1.1 Existence and Uniqueness Given an open intervalI== (a,b)( ∞ a<b +∞), and a generic weight func- tion ω such that − ≤ ≤ ω(x)>6, x I and ω L 1(I), (..1# ∀ ∈ ∈ / two functionsf andg are said to be orthogonal to each other inL ω (a,b) or orthog- onal with respect to ω if b (f,g) ω == f(x)g(x) ω(x)dx=6. �a <n algebraic polynomial of degreen is denoted by n n 1 pn(x)=k nx +k n 1x − +...+k 1x+k 6,k n =6, (../# − � where k i are real constants, andk n is the leading coef*cient ofp n. { } ∞ < se%uence of polynomials p n n=6 with deg(pn) =n is said to be orthogonal / { } inL ω (a,b) if b (pn,p m)ω = pn(x)pm(x)ω(x)dx= γnδmn, (...# �a / where the constant γn = p n is non ero, and δmn is the >ronec+er delta. � �ω Throughout this section, p n denotes a se%uence of polynomials orthogonal { } with respect to the weight function ω, andp n is of degreen. 3enote byP n the set of all algebraic polynomials of degree n, namely, ≤ / n Pn == span 1,x,x ,...,x . (..0# :y a dimension argument, � � Pn = span p 6,p 1,...,p n . (..1# { } < direct conse%uence is the following. Lemma 3.1.p n 1 is orthogonal to any polynomial q P n. + ∈ Proof Than+sto(..1#, for anyq P n, we can write ∈ q=b n pn +b n 1 pn 1 +...+b 6 p6. − − (ence, (pn+1,q) ω =b n(pn+1,p n)ω +b n 1(pn+1,p n 1)ω +...+b 6(pn+1,p 6)ω =6, − − where we have used the orthogonality (...#. �� (ereafter, we denote the monic polynomial corresponding top n by n (n) n 1 (n) ?pn(x)==p n(x)/kn =x +a n 1x − +...+a 6 . (..@# − ..1 Orthogonal ;olynomials 0! ,e show below that for any given weight function ω(x) de*ned in(a,b), there e)ists a uni%ue family of monic orthogonal polynomials generated by a three-term recurrence formula. Theorem 3.1. !or any given positive #eight function ω L 1(I), there exists a ∈ $niq$e seq$ence of monic orthogonal polynomials ?pn #ith deg(?pn) = n, #hich can be constr$cted as follo#s { } ?p6 =1,?p 1 =x α6, − (.."# ?pn+1 = (x αn)?pn βn ?pn 1,n 1, − − − ≥ #here (x?pn,?pn)ω αn = / ,n 6, (..5a# ?pn ω ≥ � �/ ?pn ω βn = � � / ,n 1. (..5b# ?pn 1 ω ≥ � − � Proof It is clear that the *rst two polynomials are ?p6(x) 1,?p 1(x)=x α6. ≡ − To determine α6, we see that(?p6,?p1)ω =6 ifandonlyif b b (x?p6,?p6)ω α6 = ω(x)xdx ω(x)dx= / , a a ?p6 ω � �� � � where the denominator is positive due to (..1#. ,e proceed with the proof by using an induction argument. <ssuming that by a similar construction, we have derived a se%uence of monic orthogonal polynomials n ?pk . Ae)t,wesee+ ?pn 1 of the form { }k=6 + n / − (n) ?pn+1 =x?pn αn ?pn βn ?pn 1 ∑ γk ?pk, (..!# − − − − k=6 with αn and βn given by (..5#, and we re%uire ?pn 1,?pk =6,6 k n. (..16# + ω ≤ ≤ � � Ta+ing the inner product with ?pk on both sides of (..!#, and using the orthogo- n nality of ?pk ,we *nd that(..16# is ful*lled if and only if { }k=6 (?pn 1,?pn)ω = (x?pn,?pn)ω αn(?pn,?pn)ω =6, + − (?pn+1,?pn 1)ω = (x?pn,?pn 1)ω βn(?pn 1,?pn 1)ω =6, − − − − − n / − (n) (..11# (?pn+1,?p%)ω = (x?pn,?p%)ω ∑ γk (?pk,?p%)ω − k=6 (...# (n) / = (x?pn,?p%)ω γ ?p% =6,6 % n /. − % � �ω ≤ ≤ − 16 . Orthogonal ;olynomials and 4elated <ppro)imation 4esults Hence, the *rst e%uality implies (..5a#, and by the second one, / (x?pn,?pn 1)ω (?pn,x?pn 1)ω ?pn ω βn = −/ = −/ = � � / , ?pn 1 ω ?pn 1 ω ?pn 1 ω � − � � − � � − � where we have used the fact n 1 − (n) x?pn 1 =?pn + ∑ δk ?pk, − k=6 n and the orthogonality of ?pk k=6 to deduce the last identity. It remains to show that (n) n {/ } the coef*cients γk k=−6 in (..!# are all ero. Indeed, we derive from 'emma ..1 that { } x?pn,?p% = (?pn,x?p%)ω =6,6 % n /, ω ≤ ≤ − � � (n) which, together with the last e%uation of (..11#, implies γk 6for6 k n /, in (..!#. This completes the induction. ≡ ≤ ≤ − Ae)t, we show that the polynomial se%uence generated by (.."#B(..5# is uni%ue. ∞ For this purpose, we assume that ?qn is another se%uence of monic orthogonal { }n=6 polynomials. &ince ?pn, given by (.."#,is also monic,we have deg ?pn+1 ?qn+1 n. :y 'emma ..1, − ≤ � � (?pn 1,?pn 1 ?qn 1)ω =6,(?q n 1,?pn 1 ?qn 1)ω =6, + + − + + + − + which implies ?pn 1 ?qn 1,?pn 1 ?qn 1 =6 ?p n 1(x) ?qn 1(x) 6. + − + + − + ω ⇒ + − + ≡ This proves� the uni%ueness. � �� The above theorem provides a general three-term recurrence relation to construct orthogonal polynomials, and the constants αn and βn can be evaluated e)plicitly for the commonly used families. The three-term recurrence relation (.."# is essential for deriving other properties of orthogonal polynomials. For convenience, we *rst e)tend it to the orthogonal polynomials p n , which are not necessarily monic. { } Corollary 3.1. Let p n be a seq$ence of orthogonal polynomials #ith the leading { } coef&cient kn =6. 'hen #e ha"e � pn+1 = (anx b n)pn c n pn 1,n 6, (..1/# − − − ≥ #ith p 1 ==6, p 6 =k 6 and − kn+1 an = , (..1.a# kn kn+1 (xpn,p n)ω bn = / , (..1.b# kn pn ω � � / kn 1kn+1 pn ω cn = − / � � / . (..1.c# kn pn 1 ω � − � ..1 Orthogonal ;olynomials 11 Proof This result follows directly from Theorem ..1 by inserting ?pl =p l/kl with l=n 1,n,n+ 1 into ( .."# and (..5#. − �� The orthogonal polynomials p n with leading coef*cients k n are uni%uely de- termined by (..1/#B(..1.#. Interestingly,{ } the following result,{ which} can be viewed as the converse of $orollary ..1, also holds. ,e leave its proof as an e)ercise (see ;roblem ..1#. Corollary 3.2. Let k n =6 be a seq$ence of real n$mbers 'he three(term re( c$rrence relation( {..1/#�)(..1.} # generates a seq$ence of polynomials satisfying the properties: the leading coef&cientof pn is kn and deg(pn) =nC • pn are orthogonal #ith respect to the #eight f$nction ω(x)C •{ }/ theL (norm of pn is gi"en by • ω / γn = p n = (a6/an)c1c/ ...c nγ6,n 6, (..10# � �ω ≥ / b #here γ6 =k 6 a ω(x)dx. <n important conse%uence� of the three-term recurrence formula (..1/#B(..1.# is the well-+nown *hristoff(+arbo$x form$la. Corollary 3.3. Let p n be a seq$ence of orthogonalpolynomials #ith deg(pn) =n Then, { } n / pn+1(x)pn(y) p n(x)pn+1(y) kn+1 pn ω − = ∑ � �/ p %(x)p %(y), (..11# x y kn p % − %=6 � �ω and n / kn+1 pn ω / pn� +1(x)pn(x) p n� (x)pn+1(x)= ∑ � �/ p % (x).

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    23 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us