221B Lecture Notes

Notes on Spherical Bessel Functions

1 Definitions
We would like to solve the free Schrödinger equation

h̄2 1 d2 h̄2 k 2
" #
l(l + 1)
− r− R(r) = R(r). (1)
2m r dr2 r2 2m

R(r) is the radial wave function ψ(~x) = R(r)Ylm (θ, φ). By factoring out
h̄2 /2m and defining ρ = kr, we find the equation

1 d2
" #
l(l + 1)
ρ − + 1 R(ρ) = 0. (2)
ρ dρ2 ρ2
The solutions to this equation are spherical Bessel functions. Due to some
reason, I don’t see the integral representations I use below in books on math-
emtical formulae, but I believe they are right.
The behavior at the origin can be studied by power expansion. Assuming
R ∝ ρn , and collecting terms of the lowest power in ρ, we get

n(n + 1) − l(l + 1) = 0. (3)

There are two solutions,

n=l or − l − 1. (4)

The first solution gives a positive power, and hence a regular solution at the
origin, while the second a negative power, and hence a singular solution at
the origin.
It is easy to check that the following integral representations solve the
above equation Eq. (2):

(1) (ρ/2)l Z i∞ iρt

hl (ρ) =− e (1 − t2 )l dt, (5)
l! +1

and
(2) (ρ/2)l Z i∞ iρt
hl (ρ) = e (1 − t2 )l dt. (6)
l! −1

1
By acting the derivatives in Eq. (2), one finds
1 d2
" #
l(l + 1) (1)
2
ρ− 2
+ 1 hl (ρ)
ρ dρ ρ
l Z i∞
" #
(ρ/2) 2 l l(l + 1) 2(l + 1)it 2 l(l + 1)
= − (1 − t ) + −t − + 1 dt
l! ±1 ρ2 ρ ρ2
(ρ/2)l 1 Z i∞ d h iρt i
= − e (1 − t2 )l+1 dt. (7)
l! iρ ±1 dt
Therefore only boundary values contribute, which vanish both at t = 1 and
(2)
t = i∞ for ρ = kr > 0. The same holds for hl (ρ).
(1)∗ (2)
One can also easily see that hl (ρ) = hl (ρ∗ ) by taking the complex
conjugate of the expression Eq. (5) and changing the variable from t to −t.
The integral representation Eq. (5) can be expanded in powers of 1/ρ.
(1)
For instance, for hl , we change the variable from t to x by t = 1 + ix, and
find
(ρ/2)l Z ∞ iρ(1+ix) l x l
 
(1)
hl (ρ) = − e x (−2i)l 1 − idx
l! 0 2i
l
(ρ/2)l iρ Z ∞
x k l
 
l −xρ
X
= −i e (−2i) l Ck e − x dx
l! k=0 0 2i
l
eiρ X (−i)l−k (l + k)! 1
= −i . (8)
ρ k=0 2k k!(l − k)! ρk
Similarly, we find
l
(2) e−iρ X il−k (l + k)! 1
hl (ρ) =i . (9)
ρ k=0 2k k!(l − k)! ρk
(1,2)
Therefore both hl are singular at ρ = 0 with power ρ−l−1 .
(1) (2)
The combination jl (ρ) = (hl + hl )/2 is regular at ρ = 0. This can be
(2)
seen easily as follows. Because hl is an integral from t = −1 to i∞, while
(1)
hl from t = +1 to i∞, the differencd between the two corresponds to an
integral from t = −1 to t = i∞ and coming back to t = +1. Because the
integrand does not have a pole, this contour can be deformed to a straight
integral from t = −1 to +1. Therefore,
1 (ρ/2)l Z 1 iρt
jl (ρ) = e (1 − t2 )l dt. (10)
2 l! −1

2
In this expression, ρ → 0 can be taken without any problems in the integral
(1)
and hence jl ∝ ρl , i.e., regular. The other linear combination nl = (hl −
(2)
hl )/2i is of course singular at ρ = 0.1 Note that
(1)
hl (ρ) = jl (ρ) + i nl (ρ) (11)
is analogous to
eiρ = cos ρ + i sin ρ. (12)
It is useful to see some examples for low l.
2
j0 = sinρ ρ , j1 = sin
ρ2
ρ
− cosρ ρ , j2 = 3−ρ
ρ3
sin ρ − ρ32 cos ρ,
2
n0 = − cosρ ρ , n1 = − cos ρ
ρ2
− sinρ ρ , n2 = − 3−ρ
ρ 3 cos ρ − ρ32 sin ρ,
(1) iρ (1)

(1)
 2
 (13)
h0 = −i eρ , h1 = −i ρ12 − ρi eiρ h2 = −i 3−ρ ρ 3 − 3i
ρ 2 eiρ .
−iρ 3−ρ2
   
(2) (2) (2)
h0 = i e ρ , h1 = i 1
ρ2
+ i
ρ
e−iρ h2 = i ρ3
+ 3i
ρ2
e−iρ .

2 Power Series Expansion

Eq. (5) can be used to obtain the power series expansion. We first split the
integration region into two parts,
(ρ/2)l Z i∞ iρt (ρ/2)l
Z i∞ Z 1 
(1) 2 l
hl (ρ) =− e (1 − t ) dt = − − eiρt (1 − t2 )l dt.
l! +1 l! 0 0
(14)
The first term can be expanded in a power series by a change of variable,
t = iτ /ρ,
!l
(ρ/2)l Z ∞ −τ τ 2 idτ
the first term = − e 1+ 2
l! 0 ρ ρ
Z ∞
1
= −i l l+1 e−τ (τ 2 + ρ2 )l dτ
l!2 ρ 0
Z ∞ l
1 −τ
X
2n 2l−2n
= −i l l+1 e l Cn ρ τ dτ
l!2 ρ 0 n=l
l
1 l!
ρ2n Γ(2l − 2n + 1)
X
= −i l l+1
l!2 ρ n=l n!(l − n)!
1
Note that my notation for nl differs from Sakurai’s by a sign as seen in Eq. (7.6.52)
on page 409. I’m sorry for that, but I stick with my convention, which was taken from
Messiah.

3
 l
1 X (2l − 2n)! 2n
= −i l l+1
ρ . (15)
2 ρ n=l n!(l − n)!
On the other hand, the second term can be expanded as
(ρ/2)l Z 1 iρt
the second term = e (1 − t2 )l dt
l! 0
∞ n
(ρ/2)l Z 1 X i nn
= ρ t (1 − t2 )l dt
l! 0 n=0 n!
∞ n Z 1
(ρ/2)l X i
= ρn tn (1 − t2 )l dt
l! n=0 n! 0
∞ n
(ρ/2)l X i n Z 1 (n−1)/2 1
= ρ x (1 − x)l dx
l! n=0 n! 0 2
∞ n
1 (ρ/2)l X i n Γ( n2 + 12 )Γ(l + 1)
= ρ . (16)
2 l! n=0 n! Γ( n2 + l + 32 )
At this point, it is useful to separate the sum to even n = 2k and odd
n = 2k + 1,
the second term
∞ ∞
(−1)k 2k Γ(k + 21 ) i(−1)k 2k+1 Γ(k + 1)
!
1 ρ l X
  X
= ρ + ρ
2 2 k=0 (2k)! Γ(k + l + 32 ) k=0 (2k + 1)! Γ(k + l + 2)
∞ ∞
ρl (−1)k Γ(k + 21 ) i(−1)k 2k+1
!
X
2k
X k!
= l+1 ρ 3 + ρ
2 k=0 (2k)! Γ(k + l + 2 ) k=0 (2k + 1)! (k + l + 1)!
(17)
(1)
Because hl (ρ) = jl (ρ) + inl (ρ), we find
∞
ρl X (−1)k Γ(k + 21 ) 2k
jl (ρ) = ρ (18)
2l+1 k=0 (2k)! Γ(k + l + 32 )
l ∞
1 X (2l − 2n)! 2n ρl X (−1)k k!
nl (ρ) = − l l+1 ρ + l+1 ρ2k+1 .
2 ρ n=l n!(l − n)! 2 k=0 (2k + 1)!(k + l + 1)!
(19)
The expression for jl can be simplified using the identity Γ(n + 12 ) =
√
(n − 12 )(n − 32 ) · · · 23 12 Γ( 12 ) = (2n−1)!!
2n
π,
∞ √
ρl X (−1)k (2k − 1)!! π/2k
jl (ρ) = l+1 √ ρ2k
2 k=0 (2k)! (2k + 2l + 1)!! π/2k+l+1

4
 ∞
(−1)k 1
= ρl ρ2k
X

k=0 (2k)! (2k + 2l + 1)(2k + 2l − 1) · · · (2k + 1)

∞
(−1)k (2k + 2l)(2k + 2l − 2) · · · (2k + 2)(2k)! 2k
= ρl
X
ρ
k=0 (2k)! (2k + 2l + 1)!
∞
(−1)k 2l (k + l)!
= ρl ρ2k
X

k=0 k!(2k + 2l + 1)!

∞
(−1)k (k + l)!
= (2ρ)l ρ2k .
X
(20)
k=0 k!(2k + 2l + 1)!

To write out the first three terms,

ρl ρ2 ρ4
" #
= 1− + − ··· . (21)
(2l + 1)!! 2(2l + 3) 8(2l + 5)(2l + 3)

It suggests that the leading term is a good approximation when ρ  2l1/2 .

Similarly, the expression for nl (ρ) can also be simplified,
l ∞
1 X (2l − 2n)! 2n ρl X (−1)k k!
nl (ρ) = − ρ + ρ2k+1
2l ρl+1 n=l n!(l − n)! 2l+1 k=0 (2k + 1)!(k + l + 1)!
l ∞
(−1)k k!
!
1 (2l − 2n)! 2n 1 X
ρ2k+2l+2
X
= − l l+1 ρ −
2ρ n=l n!(l − n)! 2 k=0 (2k + 1)!(k + l + 1)!
 
l ∞
1 X (2l − 2n)! 2n 1 X (−1)n+l (n − l − 1)! 2n 
= − l l+1 ρ + ρ
2ρ n=l n!(l − n)! 2 n=l+1 (2n − 2l − 1)!n!
 
l ∞ n+l
1 X (2l − 2n)! 2n X (−1) (n − l)! 2n 
= −  ρ + ρ
2l ρl+1 n=l n!(l − n)! n=l+1 (2n − 2l)!n!
l
1 X (−1)n Γ(2l − 2n + 1) 2n
= − ρ . (22)
2l ρl+1 n=l n! Γ(l − n + 1)

To write out the first three terms,

ρ2 ρ4
" #
(2l − 1)!!
= − 1 + + + ··· . (23)
ρl+1 2(2l − 1) 8(2l − 1)(2l − 3)

It suggests that the leading term is a good approximation when ρ  2l1/2 .

5
3 Asymptotic Behavior
(1)
Eqs. (8,9) give the asymptotic behaviors of hl for ρ → ∞:

(1) eiρ l ei(ρ−lπ/2)

hl ∼ −i (−i) = −i . (24)
ρ ρ
By taking linear combinations, we also find

sin(ρ − lπ/2)
jl ∼ , (25)
ρ
cos(ρ − lπ/2)
nl ∼ − . (26)
ρ

These expressions are good approximations when ρ  l2 . As seen in the

next section, there are better approximations when ρ ≥ l ≥ 1.

4 Large l Behavior
Starting from the integral from Eq. (10), we use the steepest descent method
to find the large l behavior. Changing the variable t = lτ ,

1 (ρ/2)l Z 1 iρt
jl (ρ) = e (1 − t2 )l dt
2 l! −1
1 (ρ/2)l Z 1/l l(log(1−l2 τ 2 )+iρτ )
= e ldτ
2 l! −1/l
√
1 (ρ/2)l Z 1/l
" q
2 2
2l(l − l2 − ρ2 )
= exp −l + l − ρ + l log
2 l! −1/l ρ2
√ √ !2 
lρ2 l2 − ρ2 l − l 2 − ρ2
− √ τ −i + O(∆τ )3  ldτ
2(l − l2 − ρ2 ) lρ
√ √ !l √ !1/2
1 (ρ/2)l −l l2 −ρ2 2l(l − l2 − ρ2 ) 2π(l − l2 − ρ2 )
' √ e e √ l
2 2πl ll e−l ρ2 lρ2 l2 − ρ2
√ √
1 √l2 −ρ2 l − l2 − ρ2
!l !1/2
l − l 2 − ρ2
= e √2 . (27)
2ρ ρ l − ρ2

This expression works very well as long as l  1 and ρ ≤ l.

6
 Starting from the integral from Eq. (5), we use the steepest descent
method to find the large l behavior. Change the variable t = ilτ ,

(1) (ρ/2)l Z i∞ iρt

hl (ρ)=− e (1 − t2 )l dt
l! +1
√2
(ρ/2)l Z ∞
" q 2l(l + l − ρ2 )
= −il exp −l − l2 − ρ2 + l log
l! −i/l ρ2
√ √ !2 
lρ2 l2 − ρ2 l + l 2 − ρ2
− √ τ− + O(∆τ )3 
2(l + l2 − ρ2 ) lρ
√ √ !l √ !1/2
− l2 −ρ2 l + l 2 − ρ2 l + l 2 − ρ2
' −ie √ . (28)
ρ ρ2 l2 − ρ2

Note that there are actually two saddle points,

√
l ± l2 − ρ2
τ= . (29)
lρ
In the above calculation, we picked the saddle point with the negative sign
with the steepest descent, while the other saddle point is what we picked for
jl (ρ). Therefore,
√ √
1 √2 2
!l !1/2
l+ l 2 − ρ2 l + l 2 − ρ2
nl (ρ) ' − e− l −ρ √2 . (30)
ρ ρ l − ρ2

This expression again works very well as long as l  1 and ρ ≤ l (but not
too close).
On the other hand, for ρ ≥ l  1, the saddle points above become
(1)
complex.
√ The contribution to the hl (ρ) is given by the saddle point τ =
l−i ρ2 −l2
lρ
, and hence
√ √
1 √2 2
!l !1/2
(1) l − i ρ2 − l 2 l − i ρ2 − l 2
hl (ρ) ' ei ρ −l √ . (31)
ρ ρ i ρ2 − l 2

This works very well as long as l  1 and ρ ≥ l (but not too close). jl (nl )
(1)
is given by the real (imaginary) part of hl (ρ). In practice, this form works
remarkably well even for l = 1.

7
 jl(ρ)
0.02
large l
l=100

0.01 exact

ρ
120 140 160 180 200

-0.01
large ρ

Figure 1: Comparison of the large ρ behavior and the large l behavior of jl (ρ)
to the exact result. The large l behavior is a very good approximation for
ρ ≥ 105 > 100 = l, while the large ρ behavior is still a poor approximation
unless ρ > O(l2 ).

(1)
It is interesting to note that this asymptotic behavior of hl (ρ) is what
you expect from the semi-classical approximation for the free-particle wave
function. The classical action for a free particle is
s s
Z r l2 q kr − l
S(r) = h̄ k 2 − 02 dr0 = h̄ (kr)2 − l2 − 2l arctan (32)
l/k r kr + l
and hence q l
√

(kr)2 −l2
l − i (kr)2 − l2
eiS(r)/h̄ = ei   , (33)
kr
which agrees with the large l behavior above except for the last factor which
comes from the lowest-order quantum correction.
When ρ ' l  1, two saddle points collide and I don’t know what to do.

5 Recursion Formulae
Starting from Eq. (5), we take the derivative

d (1) l (1) (ρ/2)l Z i∞ iρt

hl = hl − e it(1 − t2 )l dt. (34)
dρ ρ l! +1

The second term can be integrated by parts, and gives

l (1) ρ (ρ/2)l Z i∞ iρt l (1) (1)
= hl + e (1 − t2 )l+1 dt = hl − hl+1 . (35)
ρ 2 l! +1 ρ

8
 (2)
In fact, other functions jl , nl , and hl all satisfy the same relation which can
be easily checked. Referring to all of them generically as zl (ρ), we find the
recursion formula
l
zl0 = zl − zl+1 . (36)
ρ
Because zl (ρ) satisfies the differential equation Eq. 2, we can combine it
with the above recursion relation and find
d2
!
2 d l(l + 1)
0 = 2
+ − + 1 zl
dρ ρ dρ ρ2
2l + 3
= zl − zl+1 + zl+2 . (37)
ρ
Relabeling l to l − 1, we obtain
2l + 1
zl−1 + zl+1 = zl . (38)
ρ
Finally, combining the two recursion relations, we also obtain
l+1
zl0 = zl−1 − zl . (39)
ρ

6 Plane Wave Expansion

The non-trivial looking formula we used in the class
∞
eikz = (2l + 1)il jl (kr)Pl (cos θ)
X
(40)
l=0

can be obtained quite easily from the integral representation Eq. (10). The
point is that one can keep integrating it in parts. By integrating eiρt factor
and differentiating (1 − t2 )l factor, the boundary terms at t = ±1 always
vanish up to l-th time because of the (1 − t2 )l factor. Therefore,
!l
1 (ρ/2)l Z 1 1 iρt d
jl = l
e − (1 − t2 )l dt. (41)
2 l! −1 (iρ) dt
Note that the definition of the Legendre polynomials is
1 1 dl 2
Pl (t) = (t − 1)l . (42)
2l l! dtl
9
Using this definition, the spherical Bessel function can be written as
1 1 Z 1 iρt
jl = l e Pl (t)dt. (43)
2 i −1
Then we use the fact that the Legendre polynomials form a complete set of
orthogonal polynomials in the interval t ∈ [−1, 1]. Noting the normalization
Z 1
2
Pn (t)Pm (t)dt = δn,m , (44)
−1 2n + 1
q
the orthonormal basis is Pn (t) (2n + 1)/2, and hence
∞
2n + 1
Pn (t)Pn (t0 ) = δ(t − t0 ).
X
(45)
n=0 2
By multipyling Eq. (43) by Pl (t0 )(2l + 1)/2 and summing over n,
∞ Z 1 ∞
2l + 1 2l + 1 0
2il Pl (t0 )jl (ρ) = eiρt Pl (t0 )Pl (t)dt = eiρt .
X X
(46)
l=0 2 −1 n=0 2
By setting ρ = kr and t0 = cos θ, we prove Eq. (40).
If the wave vector is pointing at other directions than the positive z-
axis,
q the formula Eq. (40) needs to be generalized. Noting Yl0 (θ, φ) =
(2l + 1)/4π Pl (cos θ), we find
∞ l
i~k·~
x l
Ylm∗ (θ~k , φ~k )Ylm (θ~x , φ~x )
X X
e = 4π i jl (kr) (47)
l=0 m=−l

7 Delta-Function Normalization
An important consequence of the identity Eq. (47) is the innerproduct of two
spherical Bessel functions. We start with
Z
~ ~0
d~xeik·~x e−ik ·~x = (2π)3 δ(~k − ~k 0 ). (48)
Using Eq. (47) in the l.h.s of this equation, we find
Z
~ ~0
d~xeik·~x e−ik ·~x
Z
∗ 0 0 0
(4π)2 dΩ~x drr2 Ylm∗ (Ω~k )Ylm (Ω~x )Ylm (Ω~x )Ylm
XX
= 0 0 (Ω~ 0 )jl (kr)jl0 (k r)
k
l,m l0 ,m0
Z
(4π)2 drr2 jl (kr)jl (k 0 r)Ylm∗ (Ω~k )Ylm (Ω~k0 ).
X
= (49)
l,m

10
On the other hand, the r.h.s. of Eq. (48) is
1
(2π)3 δ(~k − ~k 0 ) = (2π)3 2 δ(k − k 0 )δ(Ω~k − Ω~k0 )
k
1
= (2π)3 2 δ(k − k 0 )δ(θ − θ0 )δ(φ − φ0 ). (50)
k sin θ
Comparing Eq. (49) and (50) and noting

Ylm∗ (Ω~k )Ylm (Ω~k0 ) = δ(Ω~k − Ω~k0 ),

X
(51)
l,m

we find ∞
Z
π
drr2 jl (kr)jl (k 0 r) = 2
δ(k − k 0 ). (52)
0 2k

8 Mathematica
In Mathematica, spherical Bessel functions are not defined but the usual
Bessel functions are. The jl (z) is obtained by
π
r
1
BesselJ[l + , z]
2z 2
and nl (z) by
π
r
1
BesselY[l + , z].
2z 2
You may actually want to use
π
r
1
PowerExpand[ BesselJ[l + , z]]
2z 2
etc to get rid of half-odd powers.

11