Bessel

What are Bessel Functions?
Why do we care?
Exp-arc explained
Results
Calculating Bessel Functions via the Exp-arc

Method
David Borwein1 , Jonathan M. Borwein2 , and O-Yeat Chan3

1 Department
of Mathematics, University of Western Ontario

of Computer Science, Dalhousie University
3 Department of Mathematics and Statistics, Dalhousie University
2 Faculty
Illinois Number Theory Fest, May 2007

D. Borwein, J. M. Borwein, O-Y. Chan
Calculating Bessel Functions via the Exp-arc Method

Why do we care?
Exp-arc explained
Results
Outline
Why do we care?
Exp-arc explained
Results

Why do we care?
Exp-arc explained
Results
Outline
Why do we care?
Exp-arc explained
Results

Why do we care?
Exp-arc explained
Results
The second order differential equation

z 2 y 00 + zy 0 + (z 2 2 )y = 0
is called Bessels Equation.
The ordinary Bessel function or order , or the Bessel function
of the first kind of order , denoted J (z), is a solution to this
differential equation.
J (z) can be represented as an ascending series
J (z) =
X
(1)k (z/2)2k +
k =0
k !(k + + 1)

Why do we care?
Exp-arc explained
Results

z 2 y 00 + zy 0 + (z 2 2 )y = 0
J (z) =
X
(1)k (z/2)2k +
k =0
k !(k + + 1)

Why do we care?
Exp-arc explained
Results

z 2 y 00 + zy 0 + (z 2 2 )y = 0
J (z) =
X
(1)k (z/2)2k +
k =0
k !(k + + 1)

Why do we care?
Exp-arc explained
Results
It is not difficult to show that for 6 Z, J (z) and J (z) are

linearly independent. Since Bessels Equation is second order,
for non-integer this pair generates all the solutions.
When = n is an integer, Jn (z) is also given by the generating
function
X
z
1
J (z)t n .
e 2 (t t ) =
n
n=
Replace t by t and we find that

Jn (z) = (1)n Jn (z).

Why do we care?
Exp-arc explained
Results

function
X
z
1
J (z)t n .
e 2 (t t ) =
n
n=

Jn (z) = (1)n Jn (z).

Why do we care?
Exp-arc explained
Results

function
X
z
1
J (z)t n .
e 2 (t t ) =
n
n=

Jn (z) = (1)n Jn (z).

Why do we care?
Exp-arc explained
Results
So to obtain the second solution to Bessels Equation at integer

order, define the Bessel function of the second kind Yn (z),
n Z, by
J (z) cos J (z)
.
Yn (z) := lim
n
sin
For general , Y (z) is defined as above without the limit.
We also have the following ascending series representation for
Yn (z), n Z.
Yn (z) =
n1
X
(n k 1)!(z/2)2k n
k!
k =0
!
X
(1)k (z/2)2k +n (Hk + Hk +n )
.
k !(n + k )!
2( + log(z/2))Jn (z)
k =0

Why do we care?
Exp-arc explained
Results

n Z, by
J (z) cos J (z)
.
Yn (z) := lim
n
sin
Yn (z), n Z.
Yn (z) =
n1
X
(n k 1)!(z/2)2k n
k!
k =0
!
X
(1)k (z/2)2k +n (Hk + Hk +n )
.
k !(n + k )!
2( + log(z/2))Jn (z)
k =0

Why do we care?
Exp-arc explained
Results

n Z, by
J (z) cos J (z)
.
Yn (z) := lim
n
sin
Yn (z), n Z.
Yn (z) =
n1
X
(n k 1)!(z/2)2k n
k!
k =0
!
X
(1)k (z/2)2k +n (Hk + Hk +n )
.
k !(n + k )!
2( + log(z/2))Jn (z)
k =0

Why do we care?
Exp-arc explained
Results
In addition to the J and Y Bessel functions, there are also the

modified Bessel functions I (z) and K (z), which are solutions
to the differential equation
z 2 y 00 + zy 0 (z 2 + 2 )y = 0.
I (z) is usually known as the Bessel function of imaginary
argument, and is related to J by
I (z) = e
i/2
J (iz) =
X
k =0
(z/2)2k +
,
k !(k + + 1)
and K (z) is given by

K (z) =
I (z) I (z)
.
2
sin

Why do we care?
Exp-arc explained
Results

z 2 y 00 + zy 0 (z 2 + 2 )y = 0.
I (z) = e
i/2
J (iz) =
X
k =0
(z/2)2k +
,
k !(k + + 1)

K (z) =
I (z) I (z)
.
2
sin

Why do we care?
Exp-arc explained
Results

z 2 y 00 + zy 0 (z 2 + 2 )y = 0.
I (z) = e
i/2
J (iz) =
X
k =0
(z/2)2k +
,
k !(k + + 1)

K (z) =
I (z) I (z)
.
2
sin

Why do we care?
Exp-arc explained
Results
As z , we have the asymptotic expansion

J (z)
X
(1)k ( 12 )2k ( 12 + )2k
cos(z
k !(2z)2k
k =0
!
X
(1)k ( 12 )2k +1 ( 12 + )2k +1
sin(z 2 4 )
,
k !(2z)2k +1
2
z
1/2
4)
k =0
as well as similar expressions for Y , I, and K .

Here the notation (a)k is the Pochhammer symbol given by
(a)k = a(a + 1) . . . (a + k 1).
Note that this is an asymptotic series, and diverges for fixed z.

Why do we care?
Exp-arc explained
Results

J (z)
X
(1)k ( 12 )2k ( 12 + )2k
cos(z
k !(2z)2k
k =0
!
X
(1)k ( 12 )2k +1 ( 12 + )2k +1
sin(z 2 4 )
,
k !(2z)2k +1
2
z
1/2
4)
k =0

(a)k = a(a + 1) . . . (a + k 1).

Why do we care?
Exp-arc explained
Results

J (z)
X
(1)k ( 12 )2k ( 12 + )2k
cos(z
k !(2z)2k
k =0
!
X
(1)k ( 12 )2k +1 ( 12 + )2k +1
sin(z 2 4 )
,
k !(2z)2k +1
2
z
1/2
4)
k =0

(a)k = a(a + 1) . . . (a + k 1).

Why do we care?
Exp-arc explained
Results
There exists, however, a Hadamard expansion that is

convergent:
I (z) =
ez
( + 12 ) 2z
X
( 12 )k 2z +k 1 t
2 e dt.
t
k !(2z)k 0
k =0
Note that the error behaves like N 1/2 when the series is
truncated after N terms.
For more information on the classical theory, see Watsons
book, A Treatise on the Theory of Bessel Functions."

Why do we care?
Exp-arc explained
Results

convergent:
I (z) =
ez
( + 12 ) 2z
X
( 12 )k 2z +k 1 t
2 e dt.
t
k !(2z)k 0
k =0

Why do we care?
Exp-arc explained
Results

convergent:
I (z) =
ez
( + 12 ) 2z
X
( 12 )k 2z +k 1 t
2 e dt.
t
k !(2z)k 0
k =0

Why do we care?
Exp-arc explained
Results
Outline
Why do we care?
Exp-arc explained
Results

Why do we care?
Exp-arc explained
Results
Bessels Equation arises as a special case of Laplaces

Equation with cylindrical symmetry.
Thus, Bessel functions occur often in the study of waves in two
dimensions.
For example, the AMOEBA water tank uses Bessel functions to
compute the amplitudes at which to drive its pistons to form
letters from standing waves.

Why do we care?
Exp-arc explained
Results

dimensions.

Why do we care?
Exp-arc explained
Results

dimensions.

Why do we care?
Exp-arc explained
Results
Applications to Number Theory

Hardy and the Circle Problem
Let r2 (n) denote the number of representations of the positive
integer n as a sum of two squares. The circle problem is to
determine the precise order of magnitude for the error term"
P(x) defined by
X
r2 (n) = x + P(x),
0nx
where the prime 0 on the summation sign on the left side

indicates that if x is an integer, only 12 r2 (x) is counted.

Why do we care?
Exp-arc explained
Results
Applications to Number Theory

Hardy and the Circle Problem
Let r2 (n) denote the number of representations of the positive
integer n as a sum of two squares. The circle problem is to
determine the precise order of magnitude for the error term"
P(x) defined by
X
r2 (n) = x + P(x),
0nx
where the prime 0 on the summation sign on the left side

indicates that if x is an integer, only 12 r2 (x) is counted.

Why do we care?
Exp-arc explained
Results
In 1915, Hardy proved that

X
0nx
r2 (n) = x +
X
n=1
r2 (n)
x 1/2
n
J1 (2 nx).
This is equivalent to the following result of Berndt and

Zaharescu
X 0
r2 (n) = x
0nx

q
q
1
3
J1 4 m(n + 4 )x
X

J1 4 m(n + 4 )x
X
q
q
.
+2 x
m(n + 41 )
m(n + 43 )
n=0 m=1

Why do we care?
Exp-arc explained
Results
In 1915, Hardy proved that

X
0nx
r2 (n) = x +
X
n=1
r2 (n)
x 1/2
n
J1 (2 nx).
This is equivalent to the following result of Berndt and

Zaharescu
X 0
r2 (n) = x
0nx

q
q
1
3
J1 4 m(n + 4 )x
X

J1 4 m(n + 4 )x
X
q
q
.
+2 x
m(n + 41 )
m(n + 43 )
n=0 m=1

Why do we care?
Exp-arc explained
Results
The BZ result is a corollary of an entry on page 335 of

Ramanujans Lost Notebook.
That entry is one of a pair of equations involving a
doubly-infinite series of Bessel functions. If we let
(
[x],
if x is not an integer,
F (x) =
1
x 2,
if x is an integer,
where, [x] is the greatest integer less than or equal to x, and
Z (z) := Y (z)
2
K (z);
Then the two entries read


Why do we care?
Exp-arc explained
Results

(
[x],
F (x) =
1
x 2,
if x is an integer,
Z (z) := Y (z)
2
K (z);


Why do we care?
Exp-arc explained
Results

(
[x],
F (x) =
1
x 2,
if x is an integer,
Z (z) := Y (z)
2
K (z);


Why do we care?
Exp-arc explained
Results
Entry
For x > 0 and 0 < < 1,
X
n=1
x
n
sin(2n) = x
1
2
1
4
cot()

p

p

J1 4 m(n + 1 )x
1 X X J1 4 m(n + )x
p
p
+
x
2
m(n + )
m(n + 1 )
m=1 n=0

Why do we care?
Exp-arc explained
Results
Entry
For x > 0 and 0 < < 1,
1
x log(2 sin())
4
n=1

p

p

Z1 4 m(n + 1 )x
1 X X Z1 4 m(n + )x
p
p
+
x
+
.
2
m(n + )
m(n + 1 )
F
x
n
cos(2n) =
m=1 n=0

Why do we care?
Exp-arc explained
Results
I wanted to evaluate a truncation of the right-hand side of the

second entry to verify it against the left-hand side. As an
example, computing the right-hand side (with one of the sums
truncated at 50 terms, the other truncated at 1000 terms, for a
total of 50,000 summands) at 28-digit precision in PARI took
nearly 4 hours.
As a comparison, a related sum where Z1 is replaced by a sine
was evaluated in 11 minutes to 10 million summands.

Why do we care?
Exp-arc explained
Results
I wanted to evaluate a truncation of the right-hand side of the

second entry to verify it against the left-hand side. As an
example, computing the right-hand side (with one of the sums
truncated at 50 terms, the other truncated at 1000 terms, for a
total of 50,000 summands) at 28-digit precision in PARI took
nearly 4 hours.
As a comparison, a related sum where Z1 is replaced by a sine
was evaluated in 11 minutes to 10 million summands.

Why do we care?
Exp-arc explained
Results
Outline
Why do we care?
Exp-arc explained
Results

Why do we care?
Exp-arc explained
Results
What is exp-arc?
In their recent paper to find asymptotics for Laguerre
polynomials with effective (explicit) error bounds, D. Borwein,
J. M. Borwein, and R. Crandall were led to consider the
following integral
Z
/2
I(p, q) :=
eiq ep cos d.
/2
A simple change of variable yields

I(p, q) = 4e
Z
0
1/ 2
cosh(2iq arcsin x)e2px
dx.
1 x2

Why do we care?
Exp-arc explained
Results
What is exp-arc?
following integral
Z
/2
I(p, q) :=
eiq ep cos d.
/2

I(p, q) = 4e
Z
0
1/ 2
dx.
1 x2

Why do we care?
Exp-arc explained
Results
What is exp-arc?
following integral
Z
/2
I(p, q) :=
eiq ep cos d.
/2

I(p, q) = 4e
Z
0
1/ 2
dx.
1 x2

Why do we care?
Exp-arc explained
Results
This integral, therefore, reduces to an integral involving

e arcsin x , or what BBC calls an exp-arc integral.
They then exploit the fact that the exp-arc function has a very
nice series expansion on (1, 1), namely
e
arcsin x
=1+
X
ck ( )x k
k =1
k!
where
c2k +1 ( ) =
k
Y
( 2 + (2j 1)2 ),
j=1
c2k =
k
Y
( 2 + (2j 2)2 ).
j=1

Why do we care?
Exp-arc explained
Results
This integral, therefore, reduces to an integral involving

e arcsin x , or what BBC calls an exp-arc integral.
They then exploit the fact that the exp-arc function has a very
nice series expansion on (1, 1), namely
e
arcsin x
=1+
X
ck ( )x k
k =1
k!
where
c2k +1 ( ) =
k
Y
( 2 + (2j 1)2 ),
j=1
c2k =
k
Y
( 2 + (2j 2)2 ).
j=1

Why do we care?
Exp-arc explained
Results
Plugging this into the expression for I(p, q) and interchanging

summation and integration we obtain
I(p, q) = 4e
X
gk (2iq)
k =0
(2k )!
Bk (p),
(1)
where
g0 := 1,
gk () :=
k
Y
(2j 1)2 + 2
for k 1,
j=1
and
Z
Bk (p) :=
1/ 2
2k 2px 2
x e
0
dx =
2k 2
epu u k 2 du.

Why do we care?
Exp-arc explained
Results

I(p, q) = 4e
X
gk (2iq)
k =0
(2k )!
Bk (p),
(1)
where
g0 := 1,
gk () :=
k
Y
(2j 1)2 + 2
for k 1,
j=1
and
Z
Bk (p) :=
1/ 2
2k 2px 2
x e
0
dx =
2k 2
epu u k 2 du.

Why do we care?
Exp-arc explained
Results

I(p, q) = 4e
X
gk (2iq)
k =0
(2k )!
Bk (p),
(1)
where
g0 := 1,
gk () :=
k
Y
(2j 1)2 + 2
for k 1,
j=1
and
Z
Bk (p) :=
1/ 2
2k 2px 2
x e
0
dx =
2k 2
epu u k 2 du.

Why do we care?
Exp-arc explained
Results
Note that
gk and Bk are rapidly computable via recursion
R1
The integral 0 epu u k 1/2 du is uniformly bounded for all
k > 0, and so the Bk decrease geometrically as 2k .
gk ()/(2k )! are bounded for fixed .
Therefore, the series for I(p, q) is geometrically convergent.

Why do we care?
Exp-arc explained
Results
Note that
R1

Why do we care?
Exp-arc explained
Results
Note that
R1

Why do we care?
Exp-arc explained
Results
Note that
R1

Why do we care?
Exp-arc explained
Results
Outline
Why do we care?
Exp-arc explained
Results

Why do we care?
Exp-arc explained
Results
Integral Representations
Z
Z
1
sin tz sinh t
J (z) =
cos(t z sin t)dt
e
dt,
0
Z
Z 0
1
1
sin(z sin t t)dt
(et + et cos )ez sinh t dt
Y (z) =
0
0
Z
Z
sin z cosh tt
1 z cos t
I (z) =
e
cos t dt
e
dt,
0
0
and
Z
K (z) =
0
z cosh t
1
cosh t dt =
2
ez cosh tt dt.

Why do we care?
Exp-arc explained
Results
A quick change of variable allows us to express the finite

integrals in terms of I(p, q).
For integral order, the infinite integrals in J and I disappear due
to the sin . Thus we have

1 in/2
e
I(iz, n) + ein/2 I(iz, n) ,
Jn (z) =
2
and
In (z) =

1
I(z, n) + ein I(z, n) .
2

Why do we care?
Exp-arc explained
Results
A quick change of variable allows us to express the finite

integrals in terms of I(p, q).
For integral order, the infinite integrals in J and I disappear due
to the sin . Thus we have

1 in/2
e
I(iz, n) + ein/2 I(iz, n) ,
Jn (z) =
2
and
In (z) =

1
I(z, n) + ein I(z, n) .
2

Why do we care?
Exp-arc explained
Results
To deal with the general case, we need to evaluate the infinite

integrals. A change of variables plus integration by parts gives
us
Z
Z
1 z zs arcsinh s
tz sinh t
dt =
ds.
e
e e
0
0
This is an exp-arc integral, but our expansion is only valid on
[0, 1). So what should we do?

Why do we care?
Exp-arc explained
Results
To deal with the general case, we need to evaluate the infinite

integrals. A change of variables plus integration by parts gives
us
Z
Z
1 z zs arcsinh s
tz sinh t
dt =
ds.
e
e e
0
0
This is an exp-arc integral, but our expansion is only valid on
[0, 1). So what should we do?

Why do we care?
Exp-arc explained
Results
Expand about a point other than zero!

For example, the expansion at infinity is
s e arcsinh s =
X
An ()
n=0
s2n
where A0 () = 2 and for n 1,

An () =
(1)n 2 ( + n + 1)n1
,
22n n!
This expansion is valid for |s| > 1. The coefficients satisfy

An =
( + 2n 2)( + 2n 1)
An1 .
4n(n + )

Why do we care?
Exp-arc explained
Results
Expand about a point other than zero!

For example, the expansion at infinity is
s e arcsinh s =
X
An ()
n=0
s2n
where A0 () = 2 and for n 1,

An () =
(1)n 2 ( + n + 1)n1
,
22n n!
This expansion is valid for |s| > 1. The coefficients satisfy

An =
( + 2n 2)( + 2n 1)
An1 .
4n(n + )

Why do we care?
Exp-arc explained
Results
How about at other points?

No nice formula for coefficients, but thats okay.
For fixed k , we have the expansion
e arcsinh(k +s) =
X
an (k , )
n=0
n!
sn
where for n 0,
an+2 =

1 2
2
(
n
)a
k
(2n
+
1)a
n
n+1 ,
k2 + 1
and
a0 = (k +
p
k 2 + 1) ,
a1 =
a0
k2 + 1

Why do we care?
Exp-arc explained
Results

e arcsinh(k +s) =
X
an (k , )
n=0
n!
sn
where for n 0,
an+2 =

1 2
2
(
n
)a
k
(2n
+
1)a
n
n+1 ,
k2 + 1
and
a0 = (k +
p
k 2 + 1) ,
a1 =
a0
k2 + 1

Why do we care?
Exp-arc explained
Results

e arcsinh(k +s) =
X
an (k , )
n=0
n!
sn
where for n 0,
an+2 =

1 2
2
(
n
)a
k
(2n
+
1)a
n
n+1 ,
k2 + 1
and
a0 = (k +
p
k 2 + 1) ,
a1 =
a0
k2 + 1

Why do we care?
Exp-arc explained
Results
Thus for any positive integer N, we have

Z
zs arcsinh s
ds =
X
n=0
X
an (0, )
an (k , )
n (z) + n (z)
ekz
n!
n!
k =1

+ An ()Gn (N + 12 , z, ) ,
where
Z
1/2
n (z) :=
ezs sn ds =
1/2
n (z) :=
ezs sn ds =
1/2
ez/2 n
+ n1 (z),
2n z
z
ez/2 n
ez/2
+ n1 (z),
(2)n z
2n z
z

Why do we care?
Exp-arc explained
Results
and
Gn (N, z, ) :=
=
eNz
N 2n+1
eNzs (1 + s)2n ds
( + 2n 1)( + 2n 2)
Nz

e
(2(n z 1) + )
2
+ z Gn1 (N, z, ) .
N 2n+1

Why do we care?
Exp-arc explained
Results
Key points:
All the summands are easily computable via recursion
The recursions only involve elementary operations. The
initial conditions B0 (from I) and G0 each require one
evaluation of incomplete gamma, which can be done via
continued fraction.
Each series converges geometrically, like 2N (as opposed
to the Hadamard expansion for I , which is like N )
Can choose N large to avoid the An Gn sum if we want. In
this way, we can pre-compute summands involving only
and summands involving only z for one-z many- or
one- many-z evaluations.

Why do we care?
Exp-arc explained
Results
Key points:
continued fraction.

Why do we care?
Exp-arc explained
Results
Key points:
continued fraction.

Why do we care?
Exp-arc explained
Results
Key points:
continued fraction.

Bessel

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bessel

Uploaded by

Copyright:

Available Formats

What are Bessel Functions?

Calculating Bessel Functions via the Exp-arc

David Borwein1 , Jonathan M. Borwein2 , and O-Yeat Chan3

of Mathematics, University of Western Ontario

Illinois Number Theory Fest, May 2007

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

What are Bessel Functions?

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

What are Bessel Functions?

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

The second order differential equation

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

The second order differential equation

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

The second order differential equation

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

It is not difficult to show that for 6 Z, J (z) and J (z) are

Replace t by t and we find that

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

It is not difficult to show that for 6 Z, J (z) and J (z) are

Replace t by t and we find that

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

It is not difficult to show that for 6 Z, J (z) and J (z) are

Replace t by t and we find that

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

So to obtain the second solution to Bessels Equation at integer

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

So to obtain the second solution to Bessels Equation at integer

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

So to obtain the second solution to Bessels Equation at integer

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

In addition to the J and Y Bessel functions, there are also the

and K (z) is given by

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?

In addition to the J and Y Bessel functions, there are also the

and K (z) is given by

D. Borwein, J. M. Borwein, O-Y. Chan

Calculating Bessel Functions via the Exp-arc Method

What are Bessel Functions?