18.

4 Chebyshev Polynomials 899

Z1
1 1
(b) hr i= r Rn L (↵r )Rn L (↵r )r 2 dr.
0
The quantity hr i is the average displacement of the electron from the nucleus, whereas
hr 1 i is the average of the reciprocal displacement.
a0 h 2 i 1
ANS. hr i = 3n L(L + 1) , hr 1 i = 2 .
2 n a0
18.3.9 Derive a recurrence formula for the hydrogen wave function expectation values:
s + 2 s+1 s +1 h 2 2
i
hr i (2s + 3)a 0 hr s
i + (2L + 1) (s + 1) a02 hr s 1 i = 0,
n2 4
with s 2L 1.
Hint. Transform Eq. (18.80) into a form analogous to Eq. (18.73). Multiply by
⇢ s+2 u 0 c⇢ s+1 u, with u = ⇢8. Adjust c to cancel terms that do not yield expecta-
tion values.

Z1
x2 p
18.3.10 Show that xne Hn (x y)d x = ⇡ n! Pn (y), where Pn is a Legendre polynomial.
1

18.4 CHEBYSHEV POLYNOMIALS

The generating function for the Legendre polynomials can be generalized to the following
form:
X1
1
= Cn(↵) (x)t n . (18.85)
(1 2xt + t 2 )↵
n=0
(↵)
The coefficients Cn (x) are known as the ultraspherical polynomials (also called
Gegenbauer polynomials). For ↵ = 1/2, we recover the Legendre polynomials; the
special cases ↵ = 0 and ↵ = 1 yield two types of Chebyshev polynomials that are the sub-
ject of this section. The primary importance of the Chebyshev polynomials is in numerical
analysis.

Type II Polynomials
(1)
With ↵ = 1 and Cn (x) written as Un (x), Eq. (18.85) gives
X 1
1
= Un (x)t n , |x| < 1, |t| < 1. (18.86)
1 2xt + t 2
n=0
These functions are called type II Chebyshev polynomials. Although these polynomials
have few applications in mathematical physics, one unusual application is in the develop-
ment of four-dimensional spherical harmonics used in angular momentum theory.
900 Chapter 18 More Special Functions

Type I Polynomials
With ↵ = 0 there is a difficulty. Indeed, our generating function reduces to the constant
1. We may avoid this problem by first differentiating Eq. (18.85) with respect to t. This
yields

X 1
↵( 2x + 2t) 1
= nCn(↵) (x) t n ,
(1 2xt + t 2 )↵+1
n=1

or

1
" #
Xn (↵)
x t Cn (x) n 1
= t . (18.87)
(1 2xt + t 2 )↵+1 2 ↵
n=1

(0)
We define Cn (x) as
(↵)
Cn (x)
Cn(0)(x) = lim . (18.88)
↵!0 ↵
The purpose of differentiating with respect to t was to get ↵ in the denominator and to
create an indeterminate form. Now multiplying Eq. (18.87) by 2t and adding 1 in the form
(1 2xt + t 2 )/(1 2xt + t 2 ), we obtain
1
1 t2 Xn
=1+2 C (0)(x)t n. (18.89)
1 2xt + t 2 2 n
n=1

We define Tn (x) as
8
< 1, n = 0,
Tn (x) = n (0) (18.90)
: Cn (x), n > 0.
2
Note the special treatment for n = 0. We will encounter a similar treatment of the n = 0
(0)
term when we study Fourier series in Chapter 19. Also, note that Cn is the limit indicated
in Eq. (18.88) and not a literal substitution of ↵ = 0 into the generating function series.
With these new labels,
1
1 t2 X
= T0 (x) + 2 Tn (x)t n , |x|  1, |t| < 1. (18.91)
1 2xt + t 2
n=1

We call Tn (x) the type I Chebyshev polynomials. Note that the notation and spelling of
the name for these functions differ from reference to reference. Here we follow the usage
of AMS-55 (Additional Readings).
 18.4 Chebyshev Polynomials 901

Recurrence Relations
Differentiating the generating function, Eq. (18.91), with respect to t and multiplying by
the denominator, 1 2xt + t 2 , we obtain
" 1
# 1
X X
t (t x) T0 (x) + 2 Tn (x)t n = (1 2xt + t 2 ) nTn (x)t n 1
n=1 n=1
1 h
X i
1
= nTn t n 2xnTn t n + nTn t n+1 ,
n=1

from which after several simplification steps we reach the recurrence relation
Tn+1 (x) 2x Tn (x) + Tn 1 (x) = 0, n > 0. (18.92)
A similar treatment of Eq. (18.86) yields the corresponding recursion relation for Un :
Un+1 (x) 2xUn (x) + Un 1 (x) = 0, n > 0. (18.93)
Using the generating functions directly for n = 0 and 1, and then applying these recur-
rence relations for the higher-order polynomials, we get Table 18.4. Plots of the Tn and Un
are presented in Figs. 18.4 and 18.5.
Differentiation of the generating functions for Tn (x) and Un (x) with respect to the vari-
able x leads to a variety of recurrence relations involving derivatives. For example, from
Eq. (18.89) we thus obtain
1
" 1
#
X X
2 0 n n
(1 2xt + t )2 Tn (x)t = 2t T0 (x) + 2 Tn (x)t ,
n=1 n=1

from which we extract the recursion formula

0
2Tn (x) = Tn+1 (x) 2x Tn0 (x) + Tn0 1 (x). (18.94)
Other useful recurrence formulas we can find in this way are
(1 x 2 )Tn0 (x) = nx Tn (x) + nTn 1 (x) (18.95)

Table 18.4 Chebyshev Polynomials: Type I (Left), Type II

(Right)

T0 = 1 U0 = 1
T1 = x U1 = 2x
T2 = 2x 2 1 U2 = 4x 2 1
T3 = 4x 3 3x U3 = 8x 3 4x
T4 = 8x 4 8x 2 + 1 U4 = 16x 4 12x 2 + 1
T5 = 16x 5 20x 3 + 5x U5 = 32x 5 32x 3 + 6x
T6 = 32x 6 48x 4 + 18x 2 1 U6 = 64x 6 80x 4 + 24x 2 1
902 Chapter 18 More Special Functions

T1(x)

x
−1 1

T3(x)

T2(x)

−1

FIGURE 18.4 The Chebyshev polynomials T1 , T2 , and T3 .

4 U3(x)

3 U2(x)

2 U1(x)

x
−1 1
−1

−2

−3

FIGURE 18.5 The Chebyshev polynomials U1 , U2 , and U3 .

and

(1 x 2 )Un0 (x) = nxUn (x) + (n + 1)Un 1 (x). (18.96)

Manipulating a variety of these formulas as in Section 15.1 for Legendre polynomials

one can eliminate the index n 1 in favor of Tn00 and establish that Tn (x), the Chebyshev
 18.4 Chebyshev Polynomials 903

polynomial type I, satisfies the ODE

(1 x 2 )Tn00 (x) x Tn0 (x) + n 2 Tn (x) = 0. (18.97)

The Chebyshev polynomial of type II, Un (x), satisfies

(1 x 2 )Un00 (x) 3xUn0 (x) + n(n + 2)Un (x) = 0. (18.98)

We could have defined the Chebyshev polynomials starting from these ODEs, but we chose
instead a development based on generating functions.
Processes similar to those used for the Chebyshev polynomials can be applied to the
general ultraspherical polynomials; the result is the ultraspherical ODE

d 2 (↵) d (↵)
(1 x 2) C (x) (2↵ + 1)x C (x) + n(n + 2↵)Cn(↵) (x) = 0. (18.99)
dx2 n dx n

Special Values
Again, from the generating functions, we can obtain the special values of various polyno-
mials:

Tn (1) = 1, Tn ( 1) = ( 1)n ,

T2n (0) = ( 1)n , T2n+1 (0) = 0;

(18.100)
Un (1) = n + 1 , Un ( 1) = ( 1)n (n + 1),
U2n (0) = ( 1)n , U2n+1 (0) = 0.

Verification of Eq. (18.100) is left to the exercises.

The polynomials Tn and Un satisfy parity relations that follow from their generating
functions with the substitutions t ! t, x ! x, which leave them invariant; these are

Tn (x) = ( 1)n Tn ( x), Un (x) = ( 1)n Un ( x). (18.101)

Rodrigues representations of Tn (x) and Un (x) are

( 1)n ⇡ 1/2 (1 x 2 )1/2 d n h i

Tn (x) = (1 x 2 )n 1/2
(18.102)
2n 0(n + 12 ) dxn

and

( 1)n (n + 1)⇡ 1/2 dn h i

Un (x) = (1 x 2 )n+1/2 . (18.103)
2n+1 0(n + 32 )(1 x 2 )1/2 d x n
904 Chapter 18 More Special Functions

Trigonometric Form
At this point in the development of the properties of the Chebyshev polynomials it
is beneficial to change variables, replacing x by cos ✓ . With x = cos ✓ and d/d x =
( 1/ sin ✓ )(d/d✓ ), we verify that
d 2 Tn d 2 Tn dTn dTn
(1 x 2) 2
= 2
cot ✓ , x Tn0 = cot ✓ .
dx d✓ d✓ d✓
Adding these terms, Eq. (18.97) becomes
d 2 Tn
+ n 2 Tn = 0, (18.104)
d✓ 2
the simple harmonic oscillator equation with solutions cos n✓ and sin n✓ . The special val-
ues (boundary conditions at x = 0 and 1) identify
Tn = cos n✓ = cos(n arccos x). (18.105)
For n 6= 0 a second linearly independent solution of Eq. (18.104) is labeled
Vn = sin n✓ = sin(n arccos x). (18.106)
The corresponding solutions of the type II Chebyshev equation, Eq. (18.98), become
sin(n + 1)✓
Un = , (18.107)
sin ✓
cos(n + 1)✓
Wn = . (18.108)
sin ✓
The two sets of solutions, type I and type II, are related by
Vn (x) = (1 x 2 )1/2 Un 1 (x), (18.109)

Wn (x) = (1 x 2) 1/2
Tn+1 (x). (18.110)
As already seen from the generating functions, Tn (x) and Un (x) are polynomials. Clearly,
Vn (x) and Wn (x) are not polynomials. From
Tn (x) + i Vn (x) = cos n✓ + i sin n✓
h in
= (cos ✓ + i sin ✓)n = x + i(1 x 2 )1/2 , |x|  1 (18.111)

we can apply the binomial theorem to obtain expansions

✓ ◆ ✓ ◆
n n 2 n n 4
Tn (x) = x n x (1 x 2 ) + x (1 x 2 )2 ··· (18.112)
2 4

and, for n > 0

p ✓ ◆ ✓ ◆
n n 1 n n 3
Vn (x) = 1 x2 x x (1 x 2) + · · · . (18.113)
1 3
 18.4 Chebyshev Polynomials 905

From the generating functions, or from the ODEs, power-series representations are
[n/2]
n X (n m 1)! 2m
Tn (x) = ( 1)m (2x)n (18.114)
2 m! (n 2m)!
m=0

for n 1, with [n/2] the integer part of n/2 and

[n/2]
X (n m)! 2m
Un (x) = ( 1)m (2x)n . (18.115)
m! (n 2m)!
m=0

Application to Numerical Analysis

An important feature of the Chebyshev polynomials Tn (x) with n > 0 is that as x is varied,
they oscillate between the extreme values Tn = +1 and Tn = 1. This behavior is readily
seen from Eq. (18.105) and is illustrated for T12 in Fig. 18.6. If a function is expanded in
the Tn and the expansion is extended sufficiently that the contributions of successive Tn are
decreasing rapidly, a good approximation to the truncation error will be proportional to the
first Tn not included in the expansion. In this approximation, there will be negligible error
at the n values of x where Tn is zero, and there will be maximum errors (all of the same
magnitude but alternating in sign) at the extrema of Tn that fall between the zeros. In that
sense, the errors satisfy a minimax principle, meaning that the maximum of the error has
been minimized by distributing it evenly into the regions between the points of negligible
error.

0.5

−1 − 0 .5 0 .5 1

−0.5

−1

FIGURE 18.6 The Chebyshev polynomial T12 .

906 Chapter 18 More Special Functions

Example 18.4.1 MINIMIZING THE MAXIMUM ERROR

Figure 18.7 shows the errors in four-term expansions of e x on the range [ 1, 1] carried
out in various ways: (a) Maclaurin series, (b) Legendre expansion, and (c) Chebyshev
expansion. The power series is optimum at the point x = 0 and the error increases with
increasing values of |x|. The orthogonal expansions produce a fit over the region [ 1, 1],
with the maximum errors occurring at x = ±1 and three intermediate values of x. How-
ever, the Legendre expansion has larger errors at ±1 than it has at the interior points, while
the Chebyshev expansion yields smaller errors at ±1 (with a concomitant increase in the
error at the other maxima) with the result that all the error maxima are comparable. This
choice approximately minimizes the maximum error.

0.006

−1 1

(b)

(c)

(a)
−0.006

FIGURE 18.7 Error in four-term approximations to e x : (a) Power series; (b) Legendre
expansion; and (c) Chebyshev expansion.

Orthogonality
If Eq. (18.97) is put into self-adjoint form (Section 8.2), we obtain w(x) = (1 x 2 ) 1/2
as a weighting factor. For Eq. (18.98) the corresponding weighting factor is (1 x 2 )+1/2 .
 18.4 Chebyshev Polynomials 907

The resulting orthogonality integrals,

8
> 0, m 6= n,
Z1 >
<
⇡
Tm (x)Tn (x)(1 x 2 ) 1/2 d x = , m = n 6 = 0, (18.116)
>2
>
1 :
⇡, m = n = 0,
8
> 0, m 6 = n,
Z1 >
<
⇡
Vm (x)Vn (x)(1 x 2 ) 1/2 d x = , m = n 6 = 0, (18.117)
>
1 :2
>
0, m = n = 0,

Z1
⇡
Um (x)Un (x)(1 x 2 )1/2 d x = mn , (18.118)
2
1

and

Z1
⇡
Wm (x)Wn (x)(1 x 2 )1/2 d x = mn , (18.119)
2
1
are a direct consequence of the Sturm-Liouville theory. The normalization values may best
be obtained by making the substitution x = cos ✓ .

Exercises
18.4.1 By evaluating the generating function for special values of x, verify the special values
Tn (1) = 1, Tn ( 1) = ( 1)n , T2n (0) = ( 1)n , T2n+1 (0) = 0.
18.4.2 By evaluating the generating function for special values of x, verify the special values
Un (1) = n + 1, Un ( 1) = ( 1)n (n + 1), U2n (0) = ( 1)n , U2n+1 (0) = 0.
18.4.3 Another Chebyshev generating function is
X 1
1 xt
= X n (x)t n , |t| < 1.
1 2xt + t 2
n=0
How is X n (x) related to Tn (x) and Un (x)?
18.4.4 Given
(1 x 2 )Un00 (x) 3xUn0 (x) + n(n + 2)Un (x) = 0,
show that Vn (x), Eq. (18.106), satisfies
(1 x 2 )Vn00 (x) x Vn0 (x) + n 2 Vn (x) = 0,
which is Chebyshev’s equation.
908 Chapter 18 More Special Functions

18.4.5 Show that the Wronskian of Tn (x) and Vn (x) is given by

n
Tn (x)Vn0 (x) Tn0 (x)Vn (x) = .
(1 x 2 )1/2
This verifies that Tn and Vn (n 6 = 0) are independent solutions of Eq. (18.97). Con-
versely, for n = 0, we do not have linear independence. What happens at n = 0? Where
is the “second” solution?
18.4.6 Show that Wn (x) = (1 x 2) 1/2 T
n+1 (x) is a solution of

(1 x 2 )Wn00 (x) 3x Wn0 (x) + n(n + 2)Wn (x) = 0.

18.4.7 Evaluate the Wronskian of Un (x) and Wn (x) = (1 x 2) 1/2 T

n+1 (x).

18.4.8 Vn (x) = (1 x 2 )1/2 Un 1 (x) is not defined for n = 0. Show that a second and inde-
pendent solution of the Chebyshev differential equation for Tn (x) (n = 0) is V0 (x) =
arccos x (or arcsin x).
18.4.9 Show that Vn (x) satisfies the same three-term recurrence relation as Tn (x), Eq. (18.92).
18.4.10 Verify the series solutions for Tn (x) and Un (x), Eqs. (18.114) and (18.115).
18.4.11 Transform the series form of Tn (x), Eq. (18.114), into an ascending power series.
n
X (n + m 1)!
ANS. T2n (x) = ( 1)n n ( 1)m (2x)2m , n 1,
(n m)!(2m)!
m=0
n
2n + 1 X ( 1)m+n (n + m)!
T2n+1 (x) = (2x)2m+1 .
2 (n m)!(2m + 1)!
m=0

18.4.12 Rewrite the series form of Un (x), Eq. (18.115), as an ascending power series.
n
X (n + m)!
ANS. U2n (x) = ( 1)n ( 1)m (2x)2m,
(n m)!(2m)!
m=0
n
X (n + m + 1)!
U2n+1 (x) = ( 1)n ( 1)m (2x)2m+1.
(n m)!(2m + 1)!
m=0

18.4.13 (a) From the differential equation for Tn (in self-adjoint form) show that

Z1
dTm (x) dTn (x)
(1 x 2 )1/2 d x = 0, m 6= n.
dx dx
1

(b) Confirm the preceding result by showing that

dTn (x)
= nUn 1 (x).
dx
18.4.14 The substitution x = 2x 0 1 converts Tn (x) into the shifted Chebyshev polynomials
Tn⇤ (x 0 ). Verify that this produces the shifted polynomials shown in Table 18.5 and that
 18.4 Chebyshev Polynomials 909

Table 18.5 Shifted Type I Chebyshev Polynomials

T0⇤ = 1
T1⇤ = 2x 1
T2⇤ = 8x 2 8x + 1
T3⇤ = 32x 3 48x 2 + 18x 1
T4⇤ = 128x 4 256x 3 + 160x 2 32x + 1
T5⇤ = 512x 5 1280x 4 + 120x 3 400x 2 + 50x 1
T6⇤ = 2048x 6 6144x 5 + 6912x 4 3584x 3 + 840x 2 72x + 1

they satisfy the orthonormality condition

Z1
mn ⇡
Tn⇤ (x 0 )Tm (x 0 )[x(1 x)] 1/2
dx = .
2 n0
0

18.4.15 The expansion of a power of x in a Chebyshev series leads to the integral

Z1
dx
Imn = x m Tn (x) p .
1 x2
1

(a) Show that this integral vanishes for m < n.

(b) Show that this integral vanishes for m + n odd.

18.4.16 Evaluate the integral

Z1
dx
Imn = x m Tn (x) p
1 x2
1

for m n and m + n even by each of two methods:

(a) Replacing Tn (x) by its Rodrigues representation.

(b) Using x = cos ✓ to transform the integral to a form with ✓ as the variable.

m! (m n 1)!!
ANS. Imn = ⇡ , m n, m + n even.
(m n)! (m + n)!!
18.4.17 Establish the following bounds, 1  x  1:
d
(a) |Un (x)|  n + 1, (b) Tn (x)  n 2 .
dx
18.4.18 (a) Show that for 1  x  1, |Vn (x)|  1.
(b) Show that Wn (x) is unbounded in 1  x  1.
910 Chapter 18 More Special Functions

18.4.19 Verify the orthogonality-normalization integrals for

(a) Tm (x), Tn (x), (b) Vm (x), Vn (x),

(c) Um (x), Un (x), (d) Wm (x), Wn (x).
Hint. All these can be converted to trigonometric integrals.

18.4.20 Show whether

(a) Tm (x) and Vn (x) are or are not orthogonal over the interval [ 1, 1] with respect
to the weighting factor (1 x 2 ) 1/2 .
(b) Um (x) and Wn (x) are or are not orthogonal over the interval [ 1, 1] with respect
to the weighting factor (1 x 2 )1/2 .

18.4.21 Derive

(a) Tn+1 (x) + Tn 1 (x) = 2x Tn (x),

(b) Tm+n (x) + Tm n (x) = 2Tm (x)Tn (x), from the “corresponding” cosine identities.

18.4.22 A number of equations relate the two types of Chebyshev polynomials. As examples
show that

Tn (x) = Un (x) xUn 1 (x)

and

(1 x 2 )Un (x) = x Tn+1 (x) Tn+2 (x).

18.4.23 Show that

d Vn (x) Tn (x)
= np
dx 1 x2

(a) using the trigonometric forms of Vn and Tn ,

(b) using the Rodrigues representation.

18.4.24 Starting with x = cos ✓ and Tn (cos ✓) = cos n✓ , expand

✓ ◆k
k ei✓ + e i✓
x =
2
and show that
 ✓ ◆ ✓ ◆
k 1 k k
x = 1
Tk (x) + Tk 2 (x) + Tk 4 + ··· ,
2k 1 2
the series in brackets terminating after the term containing T1 or T0 .
 18.5 Hypergeometric Functions 911

18.4.25 Develop the following Chebyshev expansions (for [ 1, 1]):

" 1
#
2 X
(a) (1 x 2 )1/2 = 1 2 (4s 2 1) 1 T2s (x) ,
⇡
s=1
) 1
+1, 0 < x  1 4X
(b) = ( 1)s (2s + 1) 1 T2s+1 (x).
1, 1  x < 0 ⇡
s=0

18.4.26 (a) For the interval [ 1, 1] show that

1
1 X (2s 3)!!
|x| = + ( 1)s+1 (4s + 1)P2s (x)
2 (2s + 2)!!
s=1
1
2 4X 1
= + ( 1)s+1 2 T2s (x).
⇡ ⇡ 4s 1
s=1

(b) Show that the ratio of the coefficient of T2s (x) to that of P2s (x) approaches (⇡s) 1
as s ! 1. This illustrates the relatively rapid convergence of the Chebyshev
series.
Hint. With the Legendre recurrence relations, rewrite x Pn (x) as a linear combination
of derivatives. The trigonometric substitution x = cos ✓, Tn (x) = cos n✓ is most helpful
for the Chebyshev part.
18.4.27 Show that
1
⇡2 X
=1+2 (4s 2 1) 2
.
8
s=1

Hint. Apply Parseval’s identity (or the completeness relation) to the results of
Exercise 18.4.26.
18.4.28 Show that
1
1 ⇡ 4X 1
(a) cos x= T2n+1 (x).
2 ⇡ (2n + 1)2
n=0
1
4X 1
(b) sin 1 x = T2n+1 (x).
⇡ (2n + 1)2
n=0

18.5 HYPERGEOMETRIC FUNCTIONS

In Chapter 7 the hypergeometric equation8
x(1 x)y 00 (x) + [c (a + b + 1)x]y 0 (x) ab y(x) = 0 (18.120)

8 This is sometimes called Gauss’ ODE. The solutions are then referred to as Gauss functions.