Professional Documents
Culture Documents
Project for the Penn State - Gttingen Summer School on Number Theory
Martin Kreh
Content
Content
1.1
1.2
2.1
Generating function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2
Special values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3
Integral representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4
2.5
Asymptotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6
2.7
Integral Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
References
Martin Kreh
21
bk xk .
k=0
Then we have
0
xy (x) = nx
bk x + x
bk xk + 2nxn
k=0
x y(x) = x
kbk xk + xn
k=0
We further have
2
kbk xk
k=0
k=0
and
k(k 1)xk .
k=0
bk2 xk
k=2
bk xk + 2nxn + nxn
k=0
+ xn
X
k=0
Martin Kreh
kbk xk + xn
X
k=2
bk xk
k=0
X
2 n
bk2 xk n x
bk xk
k=0
= xn
= xn
X
(n(n 1)bk + 2nkbk + k(k 1)bk + nbk + kbk + bk2 n2 bk )xk
k=0
k=0
2n is not a negative whole number. But in this case the recursion will also be fullfilled if
b2k1 = 0 for all k. So it remains to consider even k. Here we have no condition on b0 .
We can choose it freely and will do so later. Now if n
/ N we get
b2k =
b2k2
4k(n + k)
4k k!(n
1
2n n! .
It is convenient to choose b0 =
y(x) =
b0
.
+ k)(n + k 1) (n + 1)
So together we get
x n X
k=0
(1)k x 2k
.
k!(n + k)! 2
This is convergent due to the quotient criterion, therefore we have indeed constructed a
solution for the Bessel equation.
We also want to construct a solution for complex order . Here we need to exchane
the factorial with the Gamma function. Then we get
J (x) =
x X
x 2k
(1)k
(k + 1)( + k + 1) 2
k=0
Martin Kreh
x n X
k=n
x n X
k=0
x 2k
(1)k
(k + 1)(n + k + 1) 2
x 2n+2k
(1)n+k
(n + k + 1)(k + 1) 2
= (1)n Jn (x)
and of course this solves the Bessel equation.
Now we want to determine a linearly independent solution. First, we describe how J
behaves if x 0. From the power series expansion, we immediately get
0,
<() > 0
lim J (x) = 1,
.
=0
x0
, <() < 0,
/Z
This means that J and J are two linearly independent solutions if
/ Z. If Z,
they are linearly dependent. Since (1)n = cos n, the function J (z) cos J (z) is
a solution of the Bessel equation, which vanishes if n N0 . Therefore, we define
Y (x) =
for
/ Z. Then for n Z the limit
Yn (x) := lim Y (x)
n
exists, as can be shown by LHospitals rule, and J and Y are two linearly independent
solutions of the Bessel equation for all C. This can e.g. be shown be computing the
Wronskian determinant.
Martin Kreh
x X
k=0
x 2k
1
(k + 1)( + 1 + k) 2
As before, we search for a second, linearly independent solution. This will, for
/ Z, be
given by
K (x) =
I (x) I (x)
2
sin()
Martin Kreh
e 2 (zz
1 )
Jn (x)z n
(2.1)
n=
x
i.e. e 2 (zz
1 )
Proof: We have
x
x 1
X
( x2 )m m X (1)k ( x2 )k k
z
z
m!
k!
m=0
k=0
x m+k
k
X
X
(1) ( 2 )
n
=
z
m!k!
mk=n
n=
e 2 z e 2 z =
m,k0
=
=
n=
k=0
(1)k x 2k x n
(n + k)!k! 2
2
!
zn
Jn (x)z n .
n=
q.e.d.
We can use this generating function to prove some standard results.
Lemma 2.2 We have
cos x = J0 (x) + 2
n=1
sin(x) = 2
n=0
Martin Kreh
1 = J0 (x) + 2
J2n (x).
n=1
ix sin
Jn (x)e
in
n=
so
cos(x sin ) = J0 (x) + 2
Jn (x)(cos(n)+i sin(n))
n=
n=1
and
sin(x sin ) = 2
n=0
and = 0, respectively.
q.e.d.
Jn (x)z n =
n=
X
n=
Jn (x)z n =
Jn (x)z n
n=
and comparing coefficients gives the result. The second equality follows analog with the
change of variable z z 1 .
q.e.d.
(2.2)
Proof: The first statement follows, when we differentiate equation (2.1) with respect to
x:
Martin Kreh
Jn (x)z n =
n=
1
2
1
z
1
x
1 X
1 X
e 2 (z z ) = z
Jn (x)z n
Jn (x)z n
2 n=
2z n=
X
1
=
(Jn1 (x) Jn+1 (x))z n .
2
n=
q.e.d.
With the last two lemmata we can describe some integrals with Bessel functions via a
different Bessel function.
Lemma 2.5 For any n Z we have
Z
x
n+1
Jn (x) dx = x
n+1
Z
Jn+1 (x)
and
Proof: The first equation follows directly from (2.2). For the second one, we use (2.2)
and lemma 2.3 to get
Z
xn+1 Jn (x) dx =
P
2
We have already seen, that J02 (x) + 2
n=1 Jn (x) = 1. With a bit more work, we can
P
also deduce a result for
Jn+m (x)Jn (x) with m 6= 0.
Lemma 2.6 We have for any m 6= 0
J02 (x) + 2
Jn2 (x) = 1
and
n=
n=1
Martin Kreh
Jk (x)z k
Jn (x)z n
n=
k=
m=
n=
!
Jn+m (x)Jn (x) z m
Jn2 (x)
n=
!
Jn+m (x)Jn (x) z n .
n=
mZ
m6=0
Jn2 (x)
J02 (x)
+2
n=
Jn2 (x)
n=1
and
0=
n=
for all m 6= 0.
q.e.d.
Jn (x) = 1.
nZ
q.e.d.
Jk (x)Jnk (y)
kZ
for all n Z.
Proof: This follow through
1
1
Jn (x + y) = exp
(x + y)(t t )
2
nZ
1
1
= exp
x(t t1 ) exp
y(t t1 )
2
2
X
X
=
Jk (x)tk
Jm (y)tm
X
mZ
kZ
!
=
X X
nZ
Jk (x)Jnk (y) tn
kZ
q.e.d.
Most of the results hold (in a similar form) also for the other three Bessel functions.
Since the proofs are anlog, we will only state the results.
Theorem 2.9 (Results for Jn (x), Yn (x) and Kn (x))
Martin Kreh
9
x
e 2 (z+z
1 )
1 )
, i.e.
In (x)z n .
n=
Indeed this and the next two sets of formulae, as well as those for Jn (x), hold also
for complex orders, as can be shown through the series representation.
We have for all n Z
d n
(x Yn (x)) = xn Yn1 (x),
dx
d n
(x In (x)) = xn In1 (x),
dx
d n
(x Kn (x)) = xn Kn1 (x).
dx
For all n Z there holds
Z
xn+1 Yn (x) dx = xn+1 Yn+1 (x),
Z
xn+1 In (x) dx = xn+1 In+1 (x),
Z
xn+1 Kn (x) dx = xn+1 Kn+1 (x),
Martin Kreh
10
and
n=
n=1
We have
Ik (x) = ex .
n=
It holds
In (x + y) =
Ik (x)Ink (x).
kZ
1
2
1. We have
r
2
sin(x),
x
r
2
cos(x),
J 1 (x) = Y 1 (x) =
2
2
x
I 1 (x) = 2x sinh(x),
2
r
2
I 1 (x) =
cosh(x),
2
x
r
x
e .
K 1 (x) = K 1 (x) =
2
2
2x
J 1 (x) = Y 1 (x) =
2
1
1
Jk+ 1 (x) =
Pk
sin(x) Qk1
cos(x) ,
2
x
x
r
2
1
1
k
Jk 1 (x) = (1)
Pk
cos(x) + Qk1
sin(x) ,
2
x
x
x
Martin Kreh
2
x
11
1
1
k
k1
x
.
Kk+ 1 (x) = Kk 1 (x) =
i Pk
sin(x) + i Qk1
e
2
2
2x
ix
ix
Proof: The first formulae follow directly from the series representations of J and I ,
the definitions of Y and K and the properties of the -function. For example, we have
x 2k
xX
(1)k
J 1 (x) =
2
2
(k + 1)(k + 32 ) 2
k=0
r
x X (1)k x 2k
=
(2k+1)!
2
2
k=0 k! k!22k+1
r
2 X (1)k 2k+1
=
x
x
(2k + 1)!
k=0
r
2
=
sin x
x
r
and
Y 1 (x) =
sin( 2 )
= J 1 (x).
2
The rest follows from the recurrence formulas of the last section.
q.e.d.
n1
i
(Jn (ix) + iYn (ix)).
2
Martin Kreh
12
sin()
cos(x sin t t) dt
Y (x) =
sin(x sin t t) dt
1
I (x) =
0
x cos t
sin()
cos(t) dt
Z
K (x) =
ex sinh(t)t dt,
ex cosh(t)t dt,
1
(z)
1
2i
Z X
Z
(x/2)
(1)k (x/2)2k tk1 t
2
e dt =
t1 etx /(4t) dt.
k!
2i
0 k=0
u1 e(x/2)(u u ) du
for some contour of the same kind. Now let u = ew . Then the contour , given as
rectangular with vertices i, i, i, + i without the right edge is the image of
a suitable (see figure 2.1). So we get
Martin Kreh
J (x) =
1
2i
13
ew ex sinh(w) dw =
1
(I1 + I2 I3 )
2i
with
Z
I1 =
I2 =
Z
I3 =
We have
Z
1
1
I1 =
eit ex sinh(it) i dt
2i
2i
Z
Z0
1
=
ei(x sin tt) dt + ei(x sin tt) dt
2
1
2
Z
cos(x sin t t) dt
0
e(i+t) ex sinh(i+t) dt
1
2i
e(i+t) ex sinh(i+t) dt
ex sinh tt (ei ei ) dt
sin()
=
ex sinh tt dt
which gives the second part. This proves the first result.
Martin Kreh
14
I1 sin()
I2
with
Z
Z
cos(x sin(t) (t)) dt
I1 = cos()
0
Z
I2 =
Since
cos() cos(x sin(t) t) = cos(x sin(t) + ( t)) + sin() sin(x sin(t) t)
and
Z
cos(x sin(t) + ( t)) dt =
we have I1 = sin()
R
0
q.e.d.
With this theorem we can again follow that Yn and Kn exist for n Z.
With the generating function, these formulae can be proven even simplier, at least for
n Z. We will show this as an example for the functions Jn .
We use
eint dt = 2n0
to get
1
Jn (x) =
2
Z
e
Martin Kreh
Z
Jk (x)
kZ
So
i(x sin n)
1
=
Z
cos(x sin n) d.
0
15
e 2 (zz
1 )
yn (x)z n
n=
Then we have
yn (x) =
x n X
k=0
1
(1)k x 2k
and yn (x) =
(n + k)!k! 2
Z
cos(x sin n) d.
0
1 )
e 2 (tt
tn+1
dt
2u
x
4u2 x2
e 4u
2
2 n+1 n+1 x du
(x)
u
Z
x2
1 x n
=
eu e 4u un1 du
2i 2
1
yn (x) =
2i
x n
2
XX
m=0 k=0
x n X
l=0
Z
(1)k x 2k 1
umkn1 du
m!k! 2
2i
(1)k
x 2k
(n + k)!k!
Z2
i )
e 2 (e e
eni
Martin Kreh
16
Z2
1
=
2
Z
cos(x sin n) d.
0
q.e.d.
From this integral representation we can then easily, using partial integration, see that
this function solves the Bessel differential equation.
2.5 Asymptotics
Now we want to use the proven integral representations to get asymptotic formulae.
Theorem 2.13 For x R, x we have
r
2
cos(x
),
x
4
2
r
2
sin(x
),
Y (x)
x
4
2
r
1 x
I (x)
e ,
2x
r
x
e .
K (x)
2x
J (x)
Proof:
J and Y :
2ei/2
(I1 (x) + I2 (x)) + O(xA )
with
Z/3
I1 (x) =
eix cos t cosh(t) dt,
0
Martin Kreh
Z/2
I2 (x) =
eix cos t cosh(t) dt.
/3
17
with (u) =
.
1u2
1/2
Z1/2
eixu
1
(u)
eixu 0 (u) du = O(x1 ),
I2 (x) =
ix
ix
0
0
ix Zx/2
2e
u
du
iu2
1
q
I1 (x) =
e cosh 2 sin
x
2x
1
0
and so
I1 (x)
2xe
ix
u2
2x
eit dt.
Using
R
0
it2
dt =
i/4
2 e
we have
r
J (x) + iY (x)
2 i(xpi/4/2)
e
,
x
R
0
integral. We split this in two integrals, on from 0 to /2 and the other from /2 to
. Since cos(t) 0 in the second interval, the second integral is bounded. For the
Z 2x
2
u
du
u2
1
q
I (x) =
e
cos 2 sin
x
2 x
1
ex
Martin Kreh
R
0
u2
2
+ O(1).
u2
4x
2.
18
K :
u2
cosh 2 sinh
2 x
R
0
u2
2
du
q
1+
.
u2
4x
2.
q.e.d.
Martin Kreh
19
1
.
1 + s2
0=
so L[J0 ](s) =
c
1+s2
1
.
1 + s2
q.e.d.
F[J0 ](s) =
2
,
1s2
0,
Proof: Since J0
2
x
|s| < 1
|s| 1
= lim
0
Martin Kreh
J0 (t)eist|t| dt
J0 (t)et(is) dt +
J0 (t)et(+is) dt
= lim
p
+p
1 + ( is)2
1 + ( + is)2
20
2
,
1s2
|s| < 1
|s| 1
0,
q.e.d.
s s 2
2s1
sin
2
2
Proof: We use the integral representation for J0 . Since 0 < <(s) < 1, we can interchange
the two occuring integrals. The substitution y = x sin(t) then yields
1
M[J0 ](s) =
Z Z
0
1
=
sin(t)
Z
dt
y s1 cos(y) dy.
Using
Z
s1
cos(t) dt = cos
s
2
Z
(s)
and
sin(t)s dt =
tan
s ( s )
2
2 ( s+1
2 )
2
2
2 ( s+1
2 )
s1
2
s
s 2
sin
.
=
2
2
M[J0 ](s) =
q.e.d.
Martin Kreh
References
21
References
[Coh07] Cohen, Henri: Number Theory Volume II: Analytic and Modern Tools. Springer,
2007
[Dav01] Davies, Brian: Integral Transforms and Their Applications. Third Edition.
Springer, 2001
[GR00] Gradshteyn, I.S. ; Ryzhik, I.M.: Table of Integrals, Series and Products.
Sixth Edition. Academic Press, 2000
[Wat22] Watson, G.N.: A Treatise on the Theory of Bessel Functions. Cambridge
University Press, 1922
Martin Kreh