Академический Документы
Профессиональный Документы
Культура Документы
Orthogonal Polynomials
1
take this possibility into account. (This is just one of the motivating reasons for
the generalization we are about to to introduce. There are other reasons which
we do not go into here.) Moreover, the normalization of the functions n (x) may
be such that there is also an n-dependent factor on the RHS, multiplying the
Kronecker delta function. (Normalization is generally a matter of convention.)
Taking these into account, the orthonormality condition reads
Z b
n (x) m (x) (x) dx = An nm (orthonormality).
a
We will refer to the function (x) as the weight function (or simply the weight),
and the positive constant An as the normalization constant. The weight (x)
is strictly positive in (a, b) (it may vanish at the end-points a and/or b). The case
(x) = 1 corresponds to the so-called Lebesgue measure in simple terms, it
means that we just integrate any function by multiplying it by dx. When (x) is
not unity but some positive function of x in (a, b) , we say that the integration is
with respect to a measure d(x) (x) dx that is different from the Lebesgue
measure: that is, we use d rather than dx, in carrying out the integration. Note
that if the primitive or indefinite integral ofR the weight function (x) exists in
explicit form, then we would have (x) = (x) dx . But (x) may not even
exist in explicit form for a general (x) , nor is it required. The orthonormality
condition can be re-written in terms of this integration measure as
Z b
n (x) m (x) d(x) = An nm .
a
2
Given a family of orthogonal polynomials satisfying the orthonormality rela-
tion above, an arbitrary function f (x) in the function space concerned can be
expanded in terms of the complete set of basis functions as
X
f (x) = fn n (x) (expansion formula).
n=0
Explicit representation: Rb
Let Mk be the kth moment of the weight function, i.e., Mk = a xk (x) dx , k =
0, 1, . . . . Then n (x) has the explicit representation
M0 M1 M2 M
n
M1 M2 M3 Mn+1
n (x) = Cn ... .. .. .. ..
. . . .
Mn1 Mn Mn+1 M2n1
x2 xn
1 x
n (x) = an xn + bn xn1 + .
Here an+1 nd bn+1 are the leading coefficient and next-to-leading coefficient in the
polynomial n+1 (x) , etc. Several other recursion relations follow from the basic
relation above, on using various other properties of the polynomials.
Darboux-Christoffel formula:
3
It can be shown that
N
n (x) n (x 0 ) N +1 (x) N (x 0 ) N (x) N +1 (x 0 )
X 1 aN
= .
n=0
An AN aN +1 x x0
d2 d
(x) 2
+ (x) + = 0
dx dx
where (x) is a polynomial that is at most a quadratic in x, (x) is a polynomial
that is at most a linear function of x, and is a constant. There is a very
elaborate theory of hypergeometric equations and their solutions. Again, we are
only interested here in their relationship with orthogonal polynomials. We shall
therefore restrict ourselves to what is relevant to our immediate purposes.
The hypergometric equation has polynomial solutions n (x) when takes
on the special values
1
= n = n 0 n(n 1) 00 ,
2
where a prime denotes the derivative with respect to x. Such solutions are called
polynomials of the hypergeometric type.
4
the polynomial concerned, called the Rodrigues formula, which expresses the
polynomial concerned as the nth derivative of a specific function of x, thus:
(const.)n dn n
n (x) = [ (x) (x)] (Rodrigues formula) ,
(x) dxn
d2 y dy
x 2
+ ( x + 1) +ny = 0.
dx dx
d2 y dy
2
2x + 2n y = 0 .
dx dx
5
(,)
(iii) Jacobi polynomials Pn (x) , where the constants , > 1 . In this
case
(1)n dn
The Rodrigues formula for Legendre polynomials is Pn (x) = [(1 x2 )n ] .
2n n! dxn
1
2nm
Z
The orthonormality relation for Legendre polynomials is Pn (x) Pm (x) dx = .
1 2n + 1
Any continuous integrable function f (x) of x (1, 1) can be expanded in
a series of Legendre polynomials, according to
X
f (x) = (2n + 1) fn Pn (x).
n=0
Note the factor (2n + 1) in the summand on the RHS this is the usual
convention adopted. The coefficients fn in the expansion are then given by
1 1
Z
fn = Pn (x) f (x) dx .
2 1
In terms of the angle = cos1 x , this means that any continuous integrable
function F () of an angle can be expanded as
X
F () = (2n + 1) Fn Pn (cos ) ,
n=0
7
In the problems that follow, all symbols have their usual meaning as defined
in the foregoing. Thus (a, b) is a general interval, {n (x)} a family of orthogonal
polynomials defined in this interval with weight function (x) (or measure d(x) ),
and so on.
F
(a) Note that (1 2xt + t2 ) = (x t) F . Use this to show that
t
n Pn (x) (2n 1) x Pn1 (x) + (n 1) Pn2 (x) = 0 .
F F
(b) Similarly, t = (x t) . Use this to show that
t x
dPn (x) dPn1 (x)
x = n Pn (x) .
dx dx
(c) Use the results above to show that
dPn (x)
(x2 1) = (n + 1) x Pn (x) + (n + 1) Pn+1 (x) .
dx
(e) Verify that
(1)n (2n)!
Pn (1) = 1 , Pn (1) = (1)n , P2n+1 (0) = 0 , P2n (0) = .
22n (n!)2
dHn (x)
(a) Show that = 2n Hn1 (x) .
dx
(b) Show that Hn+1 (x) 2x Hn (x) + 2n Hn1 (x) = 0 .
(c) Hence show that Hn (x) satisfies the Hermite equation
9
(d) Establish the Rodrigues formula for Hermite polynomials, namely,
2 dn x2
Hn (x) = (1)n ex e .
dxn
(1)n (2n)!
(e) Show that H2n (0) = .
n!
8. The generating function of the generalized Laguerre polynomials is given
by
exp [tx/(t 1)] X
+1
= Ln (x) tn .
(1 t) n=0
(n + + 1)
(a) Show that Ln (0) = .
( + 1) n!
(b) Show that L1 (x) = + 1 x .
10