Classification and Properties of the Chebyshev Equation and its

Solutions
Emily Meyer

June 8, 2018

Background
In what follows, we study the well-known Chebyshev equation – named after the Russian mathe-
matician Pafnuty Chebyshev (1821-1894), who made wide-ranging contributions to several areas of
mathematics – from the point of view of differential equations on C. This equation is

(1 − z 2 )w00 − zw0 + x2 w = 0

where x is typically a real parameter. It is a special case of a class of ODEs known as hypergeometric-
type equations, which generically have the form

d2 w dw
σ(z) 2
+ τ (z) + ηw(z) = 0
dz dz
and also contain the Bessel equation, the Legendre equation, and many other well-known and
historically significant differential equations. In fact, the Chebyshev equation is a member of a
sub-class of hypergeometric type equations which are represented by the Gegenbauer equation:

(1 − z 2 )w00 − (a + b + 1)w0 − abw = 0.

It is easy to see that the Chebyshev equation as written takes this form, with a = x and b = −x (or
vice versa). The Chebyshev equation is of particular interest in applications, because for particular
choices of the parameter x its solutions generate an orthogonal sequence of polynomials, which
satisfy min-max properties that make them very useful for least squares approximation.
This work is divided into (1) classification and properties of the equation itself, and (2) properties
and behavior of the solutions.

Equation
We begin by analyzing the Chebyshev equation and its properties, specifically by classifying it as a
regular ODE and as a hypergeometric equation and applying important known properties of these
types of differential equations.

1
Regular Singular Points
Recall the equation
(1 − z 2 )w00 − zw0 + x2 w = 0
which is equivalent to
z x2
w00 − w 0
+ w=0 (1)
1 − z2 1 − z2
for x ∈ R and z ∈ C. Note that (1) is of the form

w00 + P (z)w0 + Q(z)w = 0

where P and Q are meromorphic functions on C. By Fuchs’ Theorem, (1) is regular if and only if
P has poles of order at most 1, and Q has poles of order at most 2. It is obvious that P and Q
each have poles at z = ±1; both functions also have a pole at z = ∞. We claim all three of these
are regular points of the ODE. To show this, we prove that the coefficients P (z) and Q(z) have
poles of order at most 1 and at most 2 (respectively) at each of these three points, as follows.
First consider z = 1:
z(z − 1)
lim (z − 1)P (z) = lim
z→1 z→1 (1 − z)(1 + z)
z
= − lim
z→1 1 + z
1
=−
2
so P has a pole of order 1 at z = 1. Similarly,

x2 (z − 1)
lim (z − 1)Q(z) = lim
z→1 z→1 (1 − z)(1 + z)

x2
= − lim
z→1 1 + z
x2
=−
2
hence Q also has a pole of order 1 at z = 1.
Next we check −1 in a similar fashion:

z(z + 1)
lim (z + 1)P (z) = lim
z→−1 (1 − z)(1 + z)
z→1
z
= lim
z→−1 1 − z
1
=
2

2
and
x2 (z + 1)
lim (z + 1)Q(z) = lim
z→−1 z→−1 (1 − z)(1 + z)

x2
= lim
z→−1 1 − z
x2
=
2
so P and Q both have poles of order 1 at z = −1.
Finally we check the point at z = ∞ by making the change of coordinates
1
ξ=
z
and evaluating the singularity at ξ = 0 of the transformed equation. Applying the substitution to
the derivative results in
d d d2 4 d
2 d
= −ξ 2 ; 2
= ξ 2
+ 2ξ 3 .
dz dξ dz dξ dξ

If W is the transformation of the resulting solution w(z), so that W (ξ) = w(z), the equation in
transformed coordinates is
1/ξ x2
ξ 4 W 00 + 2ξ 3 W 0 + ξ2W 0 + W =0
(1 − 1/ξ) (1 + 1/ξ) (1 − 1/ξ)(1 + 1/ξ)

or, simplifying,
ξ 2 x2
 
001
4
ξ W + 2+ ξ3W 0 + W = 0.
1 − ξ2 1 − ξ2
Dividing this by ξ 4 , we get

x2
 
2 1
W 00 + + W0 + W = 0.
ξ ξ(1 − ξ 2 ) ξ 2 (1 − ξ 2 )

We check the poles of the new coefficients at ξ = 0:

 
2 1
lim ξ + = lim (2 + 1) = 3
ξ→0 ξ ξ(1 − ξ 2 ) ξ→0

x2
lim ξ 2 = x2
ξ→0 ξ 2 (1 − ξ 2 )
Thus, the transformed coefficients have poles of order 1 and 2 at z = 0, so ∞ is a Fuchsian point –
that is, a regular singular point – for the original equation. We conclude that (1) is a regular ODE.

3
Hypergeometric Form and Consequences
We can make several observations about the behavior of (1) based on its classification as a hyper-
geometric type equation. To make this explicit, we make an affine change of coordinates to write
the equation as the hypergeometric equation

z(1 − z)w00 + (c − (a + b + 1)z)w0 − abw = 0.

2 2
Let t = 21 (1 − z); then z = 1 − 2t, dz
d d
= − 21 dt d
, and dz 1 d
2 = 4 dt2 , so that the equation becomes (if ẇ
represents the derivative with respect to t)
 
1
t(1 − t)ẅ + − t ẇ + x2 w = 0 (2)
2

which is in the form of the hypergeometric equation, where the constants are a, b = ±x, and c = 21 .

Symmetry of z = ±1
In the form (2), it is easy to see that the solutions will have the same behavior near z = 1 and
d d d2 d2
z = −1, by
making the substitution t 7→ 1 − t. Under this transformation, dt 7→ − dt and dt2 7→ dt2 ,
and 21 − t 7→ t − 12 , so that the equation remains unchanged:

 
1
(1 − t)tẅ − t − ẇ + x2 w = 0
2
which implies that the indicial equation and the resulting series solutions will be identical at the
two poles z = ±1 (corresponding to t = 0 and t = 1).

Monodromy Group
Furthermore, expressing the equation in the form of (2) allows us to write explicitly the matrices
generating the monodromy group for this equation:
1 −1 − e2πix
 
M (γ0 ) =
0 −1
 
−1 0
M (γ1 ) =
−1 − e−2πix 1
where γ0 and γ1 are curves in the Riemann sphere winding around the singular points t = 0 (or
z = 1) and t = 1 (where z = −1), respectively. M (γ∞ ) satisfies

M (γ0 )M (γ1 )M (γ∞ ) = I2

that is, the matrix corresponding to a curve winding around all three singularities is homotopic to
the identity.

Solutions
Next, we use the indicial equation near the regular singular points of the ODE to study the solutions
to (1).

4
Indicial Equation
To find solutions we suppose that at a pole zj , w has the form w = (z − zj )ρ . In this case, the
derivatives are w0 = ρ(z − zj )ρ−1 and w00 = ρ(ρ − 1)(z − zj )ρ−2 . The leading order terms of the
differential equation will be the (z − zj )ρ−2 terms. Then, we can evaluate the limit as z → zj
assuming the solution is of this form; the limit will depend on the pole zj . For this equation, we
have the following indicial equations:
At z = 1, the equation is

z x2
ρ(ρ − 1)(z − 1)ρ−2 + ρ(z − 1)ρ−1 + (z − 1)ρ = 0
(z − 1)(z + 1) −(z − 1)(z + 1)

which, to leading order as z → 1, is

1
ρ(ρ − 1) + ρ = 0
2
so that
1
ρ = 0 or ρ = .
2
At z = −1, we have

z x2
ρ(ρ − 1)(z + 1)ρ−2 + ρ(z + 1)ρ−1 + (z + 1)ρ = 0
(z − 1)(z + 1) −(z − 1)(z + 1)

which as z → −1, has the same leading order equation

1
ρ(ρ − 1) + ρ = 0
2
so that, as above,
1
ρ = 0 or ρ = .
2
At z = ∞, we use the equation for W (ξ), and instead suppose that W = ξ m , which results in
the equation

x2
 
m−2 2 1 m−1
m(m − 1)ξ + + mξ + ξ m−2 = 0
ξ ξ(1 − ξ 2 ) ξ 2 (1 − ξ 2 )

or, as ξ → 0 taking the O(ξ m−2 ) terms,

m(m − 1) + 3m + x2 = 0

so that p
m = −1 ± 1 − x2 .
Re-substituting ξ = z1 , this results in
√
1−x2
w(z) = z 1±

or, p
ρ=1± 1 − x2 .

5
Series Solutions
Full solutions can be written in series form based on the roots of the indicial equation, specifically
∞
X
w(z) = (z − zj )ρ an z n
n=0

We begin by assuming a solution of this form, and then replacing it into the differential equation to
solve for the coefficients an at each order. This will result in recurrence relations for the coefficients.
∞
X
w= an z n = a0 + a1 z + a2 z 2 + . . .
n=0
X∞
w0 = nan z n−1 = a1 + 2a2 z + 3a3 z 2 + . . .
n=1
X∞
w00 = n(n − 1)an z n−2 = 2a3 + 6a3 z + 12a4 z 2 + . . .
n=2

Plugging this back into (1) gives an equation for each power of z; since this holds for any z, we
can find recurrence relations for the coefficients an by equating the powers of z.
The constant terms result in the equation

x2
2a2 + x2 a0 = 0 =⇒ a2 = − a0
2
while the O(z) equation is

1 − x2
6a3 − a1 + x2 a1 = 0 =⇒ a3 = a1 .
6
In fact, for the z n terms in general, the equation will be

(n + 1)(n + 2)an+2 − n(n − 1)an − nan − x2 an = 0

which results in the recurrence relation
(n − x)(n + x)
an+2 = an .
(n + 1)(n + 2)

The initial choice of a0 is arbitrary; furthermore, since an+2 depends only on an and not on
an+1 , a1 can also be chosen arbitrarily. That is, the even and odd terms form two independent
series. Thus, the full solution is a linear combination of the series of even terms (n = 2k), which is
fully determined by the choice of a0 , and the series of odd terms (n = 2k + 1), which is determined
by the choice of a1 . These series are conventionally denoted F (z) and G(z), respectively, and
explicitly they are
∞
X (x − 2(k − 1))(x − 2(k − 2)) . . . x2 . . . (x + 2(k − 2))(x + 2(k − 1)) 2k
F (z) = (−1)2k z
(2k)!
k=0

6
 ∞
X (x − (2k − 1))(x − (2k − 3)) . . . (x − 1)(x + 1) . . . (x + (2k − 3))(x + (2k − 1)) 2k+1
G(z) = (−1)2k z
(2k + 1)!
k=0

So, the full solutions corresponding to ρ = 0 near z = ±1 are linear combinations of F (z) and G(z).
Observe that if x is a nonnegative integer, then either F or G (depending on whether x is even or
odd) terminates after n = x, resulting in a polynomial solution of degree x.
Next, we consider the solution corresponding to ρ = 21 near z = 1. For this case, we assume
solutions are of the form
∞
X
1/2
w(z) = (z − 1) an z n = (z − 1)1/2 (a0 + a1 z + a2 z 2 + . . .)
n=0

Rather than rearranging the series into a power series in (z − 1), we will use the hypergeometric
(transformed) version of equation (1):
 
1
t(1 − t)ẅ + − t ẇ + x2 w = 0.
2

In the transformed coordinates, the z = 1 point becomes t = 0, but the indicial equation near t = 0
still has roots 0 and 21 . Hence, we have a series solution of the form
∞
X ∞
X
1/2 n
w(t) = t an t = an tn+1/2
n=0 n=0

The derivatives are

∞  
X 1
w0 (t) = n+ an tn−1/2
2
n=0
∞   
X 1 1
w00 (t) = n+ n− an tn−3/2
2 2
n=0

As before, we plug these back in to the ODE to obtain recurrence relations for the coefficients
an . The first few orders areas follows:

O(t−1/2 ) : a0 = a0
1 1



1/2 2 −x 2 +x
O(t ) :a1 = 3 a0
2
3 3



3/2 2 −x 2 +x
O(t ) :a2 = a1
5

In general the equation for the terms of order O(tn+1/2 ) is

         
1 1 1 3 1 3 1
− n+ n− an + n + n+ an+1 + n+ an+1 − n + an + x2 an = 0
2 2 2 2 2 2 2

7
which simplifies to the recurrence relation
1
n + 12 + x





n+ 2 −x
an+1 = an
n + 32 (n + 1)

so the series solution can be written

∞
X (n − 1/2 − x) . . . (1/2 − x)x(x + 1/2) . . . (x + n − 1/2) n+1/2
w(t) = a0 1

1

1

t ,
n=0 2 . . . n − 2 n n + 2 (n + 1)

where a0 is arbitrary. Now, replacing t with z as the independent variable, the solution is
∞  
X 1 (n − 1/2 − x) . . . (1/2 − x)x(x + 1/2) . . . (x + n − 1/2)
w(t) = a0 1

1

1

(1 − z)n+1/2 ,
n=0
2n+1/2 2 . . . n − 2 n n + 2 (n + 1)

which converges for |1 − z| < 2.

Since each solution will satisfy the differential equation, any linear combination of the above
will also form a solution. So, we can write the general form of solutions near z = ±1 as

w(z) = a0 F (z) + a1 G(z) + a2 w(z)

for F , G, and w as above.

Due to the symmetry of (1), the series solution at z = 1 will be identical. A similar method
can be applied at z = ∞, but will depend on the parameter x; this is omitted here. However, the
two solutions near ∞ are order z ±x as z → ∞.
The preceding method is sometimes referred to as the Frobenius method.

Contour Integrals
An alternate method of finding solutions to hypergeometric type equations is to use contour inte-
gration. Hypergeometric equations satisfy particular properties with respect to integration along
certain contours; specifically, by Theorem 3.1 of ([1]), the hypergeometric equation

z(1 − z)w00 + (c − (a + b + 1)z)w0 − abw = 0

has a solution in the form of the contour integral

Z
w(z) = tb−c (1 − t)c−a−1 (t − z)−b dt
γ

over the contour γ, if γ(t) : [0, 1] → C is any curve satisfying

γ(1)
tb−c+1 (1 − t)c−a (t − (γ(t)))−b−1 γ(0) = 0.

This is proven by taking derivatives of the integral and showing that with this definition for w(z),
Z
∂ h b−c+1 i
z(1 − z)w00 + (c − (a + b + 1)z)w0 − abw = −b t (1 − t)c−a (t − (γ(t)))−b−1 dt
γ dt

8
which, by assumption on γ, must be zero.
Due to the symmetry of the Gegenbauer equation, for the Chebyshev equation this integral
form of the solution is Z
2x−1
w(z) = (t2 − 1) 2 (t − z)−x dt
γ

if γ(t) satisfies
2x+1 γ(1)
(t2 − 1) 2 (t − γ(t))−x−1 γ(0) = 0.

Furthermore, if γ(t) satisfies

γ(1)
(t2 + 2tγ(t) + 1)x+1 tx−2 γ(0) = 0

then there is a solution given by

Z
w(z) = (t2 + 2tz + 1)x tx−1 dt.
γ

We omit the proof of the final integral representation; it relies on the symmetry of Gegenbauer
equations, specifically the Whipple transformation based on the mapping z → √z±z 2 −1
.
Remark. It can be shown that several of the integral solutions and the series solutions are
proportional to each other. This implies that the Frobenius solutions can be obtained, up to a nor-
malization constant, by integrating certain elementary functions over contours satisfying particular
conditions.

References
[1] J. Dereziski, “Hypergeometric Type Functions and Their Symmetries,” 2013. [Online].
Available: http://www.fuw.edu.pl/{∼}derezins/hyper-published.pdf