Bessel functions

The Bessel function Jν (z) of the first kind of order ν is defined by

∞
(z/2)ν z2 (−1)k
 
−  z ν X  z 2k
Jν (z) = 0 F1 ;− = . (1)
Γ(ν + 1) ν+1 4 2 Γ(ν + k + 1) k! 2
k=0

For ν ≥ 0 this is a solution of the Bessel differential equation

z 2 y 00 (z) + zy 0 (z) + z 2 − ν 2 y(z) = 0, ν ≥ 0.

(2)

For ν ∈
/ {0, 1, 2, . . .} we have that J−ν (z) is a second solution of the differential equation (2)
and the two solutions Jν (z) and J−ν (z) are clearly linearly independent.
For ν = n ∈ {0, 1, 2, . . .} we have
∞ ∞
 z −n X (−1)k  z 2k  z −n X (−1)k  z 2k
J−n (z) = = ,
2 Γ(−n + k + 1) k! 2 2 Γ(−n + k + 1) k! 2
k=0 k=n

since
1
=0 for k = 0, 1, 2, . . . , n − 1.
Γ(−n + k + 1)
This implies that
∞ ∞
 z −n X (−1)k  z 2k  z −n X (−1)n+k  z 2(n+k)
J−n (z) = =
2 Γ(−n + k + 1) k! 2 2 Γ(k + 1) (n + k)! 2
k=n k=0
∞
 z n X (−1)k  z 2k
= (−1)n = (−1)n Jn (z).
2 Γ(n + k + 1) k! 2
k=0

This implies that Jn (z) and J−n (z) are linearly dependent for n ∈ {0, 1, 2, . . .}.
A second linearly independent solution can be found as follows. Since (−1)n = cos nπ, we see
that Jν (z) cos νπ − J−ν (z) is a solution of the differential equation (2) which vanishes when
ν = n ∈ {0, 1, 2, . . .}. Now we define

Jν (z) cos νπ − J−ν (z)

Yν (z) := , (3)
sin νπ
where the case that ν = n ∈ {0, 1, 2, . . .} should be regarded as a limit case. By l’Hopital’s
rule we have
(−1)n ∂J−ν (z)
   
1 ∂Jν (z)
Yn (z) = lim Yν (z) = − .
ν→n π ∂ν ν=n π ∂ν ν=n

This implies that Y−n (z) = (−1)n Yn (z) for n ∈ {0, 1, 2, . . .}.
The function Yν (z) is called the Bessel function of the second kind of order ν.
Using the definition (1) we find that
∞
(−1)k ψ(n + k + 1)  z 2k
 
∂Jν (z) z   z n X
= Jn (z) ln − ,
∂ν ν=n 2 2 (n + k)! k! 2
k=0

1
where
d Γ0 (z)
ψ(z) = ln Γ(z) = .
dz Γ(z)
For ν ∈
/ {0, 1, 2, . . .} we have
∞
 z   z −ν X (−1)k ψ(−ν + k + 1)  z 2k
∂J−ν (z)
= −Jν (z) ln + .
∂ν 2 2 Γ(−ν + k + 1) k! 2
k=0

Now we use
(−1)n (−1)n
lim (z + n)Γ(z) = =⇒ Γ(z) ∼ for z → −n
z→−n n! (z + n) n!
and n
(−1)
Γ0 (z) − (z+n) 2 n! 1
ψ(z) = ∼ (−1)n =− for z → −n.
Γ(z) z+n
(z+n) n!

This implies that

ψ(z)
lim = (−1)n+1 n! for n = 0, 1, 2, . . . .
z→−n Γ(z)
Hence
∞
X (−1)k ψ(−ν + k + 1)  z 2k
lim
ν→n Γ(−ν + k + 1) k! 2
k=0
n−1 ∞
n
X (n − k − 1)!  z 2k X (−1)k ψ(−n + k + 1)  z 2k
= (−1) +
k! 2 Γ(−n + k + 1) k! 2
k=0 k=n
n−1 ∞
X (n − k − 1)!  z 2k  z 2n X (−1)k ψ(k + 1)  z 2k
= (−1)n + (−1)n .
k! 2 2 Γ(k + 1) (n + k)! 2
k=0 k=0

This implies that


∂J−ν (z)
 z   z −n n−1
X (n − k − 1)!  z 2k
= −J−n (z) ln + (−1)n
∂ν ν=n 2 2 k! 2
k=0
∞
 z n X (−1)k ψ(k + 1)  z 2k
+ (−1)n .
2 (n + k)! k! 2
k=0

Finally we use the fact that J−n (z) = (−1)n Jn (z) to conclude that

2  z  1  z −n n−1
X (n − k − 1)!  z 2k
Yn (z) = Jn (z) ln −
π 2 π 2 k! 2
k=0
∞
1  z n X (−1)k  z 2k
− [ψ(n + k + 1) + ψ(k + 1)]
π 2 (n + k)! k! 2
k=0

for n ∈ {0, 1, 2, . . .}. Compare with the theory of Frobenius for linear second differential
equations.

2
In the theory of second order linear differential equations of the form

y 00 + p(z)y 0 + q(z)y = 0,

two solutions y1 and y2 are linearly independent if and only if

 
 y1 (z) y2 (z) 
W (y1 , y2 )(z) :=   = y1 (z)y20 (z) − y10 (z)y2 (z) 6= 0.
 
 y10 (z) y20 (z) 

This determinant is called the Wronskian of the solutions y1 and y2 .

It can (easily) be shown that this determinant of Wronski satisfies the differential equation

W 0 (z) + p(z)W (z) = 0.

This result is called Abel’s theorem or the theorem of Abel-Liouville. In the case of the Bessel
differential equation we have p(z) = 1/z, which implies that
1 c
W 0 (z) + W (z) = 0 =⇒ W (y1 , y2 )(z) =
z z
for some constant c. Now we have

Theorem 1.
2 sin νπ 2
W (Jν , J−ν )(z) = − and W (Jν , Yν )(z) = . (4)
πz πz
For ν ≥ 0 this implies that Jν (z) and J−ν (z) are linearly independent if ν ∈
/ {0, 1, 2, . . .} and
that Jν (z) and Yν (z) are linearly independent for all ν ≥ 0.

Proof. Note that

W (Jν , Yν )(z) = Jν (z)Yν0 (z) − Jν0 (z)Yν (z)

J 0 (z) cos νπ − J−ν 0 (z)
Jν (z) cos νπ − J−ν (z)
= Jν (z) ν − Jν0 (z)
sin νπ sin νπ
Jν (z)J−ν0 (z) − J 0 (z)J (z)
ν −ν W (Jν , J−ν )(z)
= − =− .
sin νπ sin νπ
Now we use the definition (1) to obtain
∞ ∞
X (−1)k z ν+2k X (−1)k (ν + 2k) z ν+2k−1
Jν (z) = · ν+2k =⇒ Jν0 (z) = ·
Γ(ν + k + 1) k! 2 Γ(ν + k + 1) k! 2ν+2k
k=0 k=0

and
∞
X (−1)m z −ν+2m
J−ν (z) = · −ν+2m
Γ(−ν + m + 1) m! 2
m=0
∞
0
X (−1)m (−ν + 2m) z −ν+2m−1
=⇒ J−ν (z) = · .
Γ(−ν + m + 1) m! 2−ν+2m
m=0

3
Hence we have
0
(z) − Jν0 (z)J−ν (z)
 
z W (Jν , J−ν )(z) = z Jν (z)J−ν
∞ X∞
X (−1)k+m (−ν + 2m) z 2k+2m
= · 2k+2m
Γ(ν + k + 1)Γ(−ν + m + 1) k! m! 2
k=0 m=0
∞ X
∞
X (−1)k+m (ν + 2k) z 2k+2m
− · 2k+2m
Γ(ν + k + 1)Γ(−ν + m + 1) k! m! 2
k=0 m=0
∞ ∞
XX (−1)k+m (2ν + 2k − 2m) z 2k+2m
= − · 2k+2m .
Γ(ν + k + 1)Γ(−ν + m + 1) k! m! 2
k=0 m=0

This implies that

2ν 2ν 2 sin νπ
lim z W (Jν , J−ν )(z) = − =− =− .
z→0 Γ(ν + 1)Γ(−ν + 1) νΓ(ν)Γ(1 − ν) π

Using the definition (1) we obtain

∞ ∞
d ν d X (−1)k z 2ν+2k X (−1)k (2ν + 2k) z 2ν+2k−1
[z Jν (z)] = · ν+2k = ·
dz dz Γ(ν + k + 1) k! 2 Γ(ν + k + 1) k! 2ν+2k
k=0 k=0
∞
X (−1)k z 2ν+2k−1
= · = z ν Jν−1 (z).
Γ(ν + k) k! 2ν+2k−1
k=0

Hence we have
d ν
[z Jν (z)] = z ν Jν−1 (z) ⇐⇒ zJν0 (z) + νJν (z) = zJν−1 (z). (5)
dz
Similarly we have
∞ ∞
d  −ν  d X (−1)k z 2k X (−1)k 2k z 2k−1
z Jν (z) = · ν+2k = · ν+2k
dz dz Γ(ν + k + 1) k! 2 Γ(ν + k + 1) k! 2
k=0 k=0
∞
X (−1)k+1 z 2k+1
= · = −z −ν Jν+1 (z).
Γ(ν + k + 2) k! 2ν+2k+1
k=0

Hence we have
d  −ν
z Jν (z) = −z −ν Jν+1 (z) zJν0 (z) − νJν (z) = −zJν+1 (z).

⇐⇒ (6)
dz
Elimination of Jν0 (z) from (5) and (6) gives
2ν
Jν−1 (z) + Jν+1 (z) = Jν (z)
z
and elimination of Jν (z) from (5) and (6) gives

Jν−1 (z) − Jν+1 (z) = 2Jν0 (z).

4
 Special cases

For ν = 1/2 we have from the definition (1) by using Legendre’s duplication formula for the
gamma function
∞  x 2k r 2x X ∞ r
(−1)k (−1)k
r
x X 2k 2
J1/2 (x) = = x = sin x, x > 0
2 Γ(k + 3/2) k! 2 π (2k + 1)! πx
k=0 k=0

and for ν = −1/2 we have

r ∞ ∞
 x 2k r 2 X r
2 X (−1)k (−1)k 2k 2
J−1/2 (x) = = x = cos x, x > 0.
x Γ(k + 1/2) k! 2 πx (2k)! πx
k=0 k=0

Note that the definition (3) implies that

r r
2 2
Y1/2 (x) = −J−1/2 (x) = − cos x and Y−1/2 (x) = J1/2 (x) = sin x, x > 0.
πx πx

Integral representations

First we will prove

Theorem 2.
1  z ν Z 1
Jν (z) = √ eizt (1 − t2 )ν−1/2 dt, Re ν > −1/2. (7)
Γ(ν + 1/2) π 2 −1

Proof. We start with

Z 1 ∞ 1
(iz)n
X Z
izt 2 ν−1/2
e (1 − t ) dt = tn (1 − t2 )ν−1/2 dt.
−1 n! −1
n=0

Note that the latter integral vanishes when n is odd. For n = 2k we obtain using t2 = u
Z 1 Z 1 Z 1
du
t2k (1 − t2 )ν−1/2 dt = 2 uk (1 − u)ν−1/2 √ = uk−1/2 (1 − u)ν−1/2 du
−1 0 2 u 0
Γ(k + 1/2)Γ(ν + 1/2)
= B(k + 1/2, ν + 1/2) = .
Γ(ν + k + 1)
Now we use Legendre’s duplication formula to find that
√ √ √
π Γ(2k) π Γ(2k + 1) π (2k)!
Γ(k + 1/2) = 2k−1 · = 2k · = 2k · .
2 Γ(k) 2 Γ(k + 1) 2 k!
Hence we have
Z 1 ∞
X (iz)2k Γ(k + 1/2)Γ(ν + 1/2)
eizt (1 − t2 )ν−1/2 dt = ·
−1 (2k)! Γ(ν + k + 1)
k=0
∞
√ X (−1)k  z 2k
= Γ(ν + 1/2) π .
Γ(ν + k + 1) k! 2
k=0

This proves the theorem.

5
We also have Poisson’s integral representations
Theorem 3.
1  z ν Z π
Jν (z) = √ eiz cos θ (sin θ)2ν dθ
Γ(ν + 1/2) π 2 0
1  z ν Z π
= √ cos(z cos θ) (sin θ)2ν dθ, Re ν > −1/2.
Γ(ν + 1/2) π 2 0

Proof. Use the substitution t = cos θ to obtain

Z 1 Z 0 Z π
izt 2 ν−1/2 iz cos θ 2 ν−1/2
e (1 − t ) dt = e (1 − cos θ) · (− sin θ) dθ = eiz cos θ (sin θ)2ν dθ.
−1 π 0

Further we have
eiz cos θ = cos(z cos θ) + i sin(z cos θ)
and Z π
sin(z cos θ)(sin θ)2ν dθ = 0.
0
This shows that Poisson’s integral representations follow from the integral representation (7).

Remarks:
1. The Fourier transform is defined by
Z ∞
1
F (z) := √ e−izt f (t) dt
2π −∞

with inversion formula Z ∞

1
f (t) = √ eizt F (z) dz.
2π −∞
This implies that the Fourier transform of the function
(
(1 − t2 )ν−1/2 , |t| ≤ 1
f (t) =
0, |t| > 1
is
Γ(ν + 1/2)  z −ν
F (z) = √ Jν (z).
2 2
2. Instead of the substitution t = cos θ in (7) one can also use the substitution t = sin θ,
which leads to slightly different forms of Poisson’s integral representations. In fact we
have
1  z ν Z π/2
Jν (z) = √ eiz sin θ (cos θ)2ν dθ
Γ(ν + 1/2) π 2 −π/2

1  z ν Z π/2
= √ cos(z sin θ) (cos θ)2ν dθ
Γ(ν + 1/2) π 2 −π/2
Z π/2
2  z  ν
= √ cos(z sin θ) (cos θ)2ν dθ, Re ν > −1/2.
Γ(ν + 1/2) π 2 0

6
 Integrals of Bessel functions

The Hankel transform of a function f is defined by

Z ∞
F (s) = tf (t)Jν (st) dt
0

for functions f for which the integral converges. The inversion formula is given by
Z ∞
f (t) = sF (s)Jν (st) ds.
0

This pair of integrals is called a Hankel pair of order ν.

An example of such an integral is

Γ µ+ν
Z ∞

sν s2
 
µ−1 −ρ2 t2 2 (µ + ν)/2
t e Jν (st) dt = ν+1 · µ+ν · 1 F1 ;− 2 , Re(µ + ν) > 0.
0 2 Γ(ν + 1) ρ ν+1 4ρ
It can be shown that the integral converges for Re(µ + ν) > 0. Now we use the definition (1)
to obtain
Z ∞ ∞ Z ∞
µ−1 −ρ2 t2
X (−1)k sν+2k 2 2
t e Jν (st) dt = · ν+2k · tµ+ν+2k−1 e−ρ t dt.
0 Γ(ν + k + 1) k! 2 0
k=0

Using the substitution ρ2 t2 = u we find that

Z ∞ Z ∞
2 2 du
tµ+ν+2k−1 e−ρ t dt = ρ−µ−ν−2k+1 u(µ+ν+2k−1)/2 e−u √
0 2ρ u
Z 0∞
1 −µ−ν−2k
= ρ u(µ+ν+2k−2)/2 e−u du
2 0
 
1 −µ−ν−2k µ + ν + 2k
= ρ Γ .
2 2
Hence we have
Z ∞ ∞ µ+ν
1 −µ−ν X (−1)k Γ 2 + k


2 2 sν+2k
tµ−1 e−ρ t Jν (st) dt = ρ · ν+2k 2k
0 2 Γ(ν + k + 1) k! 2 ρ
k=0
∞
Γ µ+ν


2 sν X (−1)k ((µ + ν)/2)k s2k
= µ+ν
· ν
· · 2k 2k
2ρ Γ(ν + 1) 2 (ν + 1)k k! 2 ρ
k=0
Γ µ+ν


sν s2
 
2 (µ + ν)/2
= · · F
1 1 ; − .
2ν+1 Γ(ν + 1) ρµ+ν ν+1 4ρ2

The special case µ = ν + 2 is of special interest: in that case we have (µ + ν)/2 = ν + 1. This
implies that the 1 F1 reduces to a 0 F0 which is an exponential function. The result is
Z ∞
2 2 sν 2 2
tν+1 e−ρ t Jν (st) dt = 2 )ν+1
· e−s /4ρ , Re ν > −1.
0 (2ρ

7
 Hankel functions
(1) (2)
The functions Hν (z) and Hν (z) are defined by
Hν(1) (z) := Jν (z) + iYν (z) and Hν(2) (z) := Jν (z) − iYν (z).
These functions are called Hankel functions or Bessel functions of the third kind.
Note that these definitions imply that
(1) (2) (1) (2)
Hν (z) + Hν (z) Hν (z) − Hν (z)
Jν (z) = and Yν (z) = .
2 2i
Further we have
r r
(1) 2 2 ix
H−1/2 (x) = J−1/2 (x) + iY−1/2 (x) = (cos x + i sin x) = e , x>0
πx πx
and
r r
(2) 2 2 −ix
H−1/2 (x) = J−1/2 (x) − iY−1/2 (x) = (cos x − i sin x) = e , x > 0.
πx πx
Similarly we have
r r
(1) 2 2 ix
H1/2 (x) = J1/2 (x) + iY1/2 (x) = (sin x − i cos x) = −i e , x>0
πx πx
and r r
(2) 2 2 −ix
H1/2 (x) = J1/2 (x) − iY1/2 (x) = (sin x + i cos x) = i e , x > 0.
πx πx

Modified Bessel functions

The modified Bessel function Iν (z) of the first kind of order ν is defined by
∞
(z/2)ν z2
    X
− z ν 1  z 2k
Iν (z) := 0 F1 ; = .
Γ(ν + 1) ν+1 4 2 Γ(ν + k + 1) k! 2
k=0
For ν ≥ 0 this is a solution of the modified Bessel differential equation
z 2 y 00 (z) + zy 0 (z) − z 2 + ν 2 y(z) = 0, ν ≥ 0.

For ν ∈
/ {0, 1, 2, . . .} we have that I−ν (z) is a second solution of this differential equation and
the two solutions Iν (z) and I−ν (z) are linearly independent.
For ν = n ∈ {0, 1, 2, . . .} we have I−n (z) = In (z).
The modified Bessel function Kν (z) of the second kind of order ν is defined by
π
Kν (z) := [I−ν (z) − Iν (z)] for ν ∈/ {0, 1, 2, . . .}
2 sin νπ
and
Kn (z) = lim Kν (z) for n ∈ {0, 1, 2, . . .}.
ν→n
Now we have for x > 0
r r r
2 2 π −x
I1/2 (x) = sinh x, I−1/2 (x) = cosh x and K1/2 (x) = K−1/2 (x) = e .
πx πx 2x

8
 A generating function

The Bessel function Jn (z) of the first kind of integer order n ∈ Z can also be defined by means
of the generating function
  ∞
1 −1
X
Jn (z)tn .


exp z t−t = (8)
2 n=−∞
In fact, the series on the right-hand side is a so-called Laurent series at t = 0 for the function
at the left-hand side. Using the Taylor series for the exponential function we obtain
∞   ∞ ∞
1 zt j X (−1)k  z k
   
1 −1 zt  z X X
an tn .


exp z t−t = exp · exp − = · =
2 2 2t j! 2 k! 2t n=−∞
j=0 k=0

For n ∈ {0, 1, 2, . . .} we have

∞ ∞
 z n+k (−1)k  z k  z n X
X 1 (−1)k  z 2k
an = · = = Jn (z)
(n + k)! 2 k! 2 2 (n + k)! k! 2
k=0 k=0
and
∞
X 1  z j (−1)j+n  z n+j
a−n = · = (−1)n Jn (z) = J−n (z).
j! 2 (n + j)! 2
j=0
This proves (8).

If t = eiθ , then we have

1
eiθ − e−iθ
t − t−1 = = i sin θ
2 2  
1 −1
= eix sin θ = cos(x sin θ) + i sin(x sin θ).


=⇒ exp x t−t
2
Hence we have
∞
X ∞
X
cos(x sin θ) + i sin(x sin θ) = eix sin θ = Jn (x)einθ = Jn (x) [cos(nθ) + i sin(nθ)] .
n=−∞ n=−∞

Since J−n (x) = (−1)n Jn (x) this implies that

∞
X ∞
X
cos(x sin θ) = Jn (x) cos(nθ) = J0 (x) + 2 J2k (x) cos(2kθ)
n=−∞ k=1
and
∞
X ∞
X
sin(x sin θ) = Jn (x) sin(nθ) = 2 J2k+1 (x) sin(2k + 1)θ.
n=−∞ k=0
For θ = π/2 this implies that
∞
X ∞
X
k
cos x = J0 (x) + 2 (−1) J2k (x) and sin x = 2 (−1)k J2k+1 (x).
k=1 k=0
For θ = 0 we also have
∞
X
1 = J0 (x) + 2 J2k (x).
k=1

9
The generating function (8) can be used to prove that
∞
X
Jn (x + y) = Jk (x)Jn−k (y).
k=−∞

The proof is as follows:

∞  
X 1
Jn (x + y)tn = exp (x + y) t − t−1


n=−∞
2
   
1 −1

1 −1


= exp x t−t · exp y t−t
2 2
∞ ∞ ∞ ∞
!
X X X X
k m
= Jk (x)t · Jm (y)t = Jk (x)Jn−k (y) tn .
k=−∞ m=−∞ n=−∞ k=−∞

We can also find an integral representation for the Bessel function Jn (x) of the first kind of
integral order n starting from the generating function
∞
X
eix sin θ = Jn (x)einθ .
n=−∞

We use the orthogonality property of the exponential function, id est

(
Z π 0, k ∈ Z, k 6= 0
eikθ dθ =
−π 2π, k = 0.

Hence we have
Z π ∞
X Z π
−inθ
ix sin θ
e ·e dθ = Jk (x) eikθ · e−inθ dθ = 2π · Jn (x).
−π k=−∞ −π

This implies that

Z π Z π
1 i(x sin θ−nθ) 1
Jn (x) = e dθ = cos (x sin θ − nθ) dθ.
2π −π π 0

The special case n = 0 reads

Z π Z 1
1 2 cos(xt)
J0 (x) = cos (x sin θ) dθ = √ dt.
π 0 π 0 1 − t2

10