221A Lecture Notes

Spherical Harmonics 1 Oribtal Angular Momentum

The orbital angular momentum operator is given just as in the classical mechanics, L = x p. (1) From this denition and the canonical commutation relation between the position and momentum operators, it is easy to verify the commutation relation among the components of the angular momentum, [Li , Lj ] = ih
ijk Lk .

(2)

When using the position representation, the action of the angular momentum on any state is given by a dierential operator h h x|L| = x x| = x (x). i i Loosely, we can write h L = x , (4) i but you have to be careful on what this dierential operator is acting on. If you act on a position ket instead of a bra, h L|x = x |x . i (5) (3)

Whenever I write the orbital angular momentum operator as a dierential operator in this note, it is understood that it acts on a position bra instead of ket. With this caveat in mind, we can rewrite the orbital angular momentum operator in the polar coordinates. Following the usual denitions x = r sin cos , y = r sin sin , z = r cos , 1

(6)

we can rewrite the derivatives using the chain rule,

r x r y r z x r = x y z y x y z z sin cos sin sin cos x = r cos cos r cos sin r sin y . r sin sin r sin cos 0 z

(7)

We invert the matrix and nd x r sin2 cos sin cos cos sin r 1 2 cos y = r sin sin sin cos sin . (8) r sin z r sin cos sin2 0 Now the orbital angular momentum operators can be written in terms of spherical coordinates, h sin cot cos , i h Ly = cos cot sin , i h . Lz = i It is useful to take the combinations h L = ei i cot . i Lx = Finally, 1 L = h sin sin The free-particle Hamiltonian is
2 2

(9) (10) (11)

(12)

1 2 + . sin2 2 2 2 + 2 r r r L2 . 2mr2

(13)

h 2 h 2 2 p2 2 2 + z )= = (x + y 2m 2m 2m

(14)

Looking at the expression for Lz , you can see that it is of the same form as the momentum operator of a particle on a circle, whose eigenvalues are quantized as nh for n Z. Therefore, Lz is also quantized as mh for m Z. An immediate consequence is that no half-odd values are allowed for Lz , and hence half-odd j cannot be obtained for orbital angular momentum. Only integer j is possible. 2

Spherical Harmonics

Now we look for eigenstates of L2 and Lz to nd the Hilbert space for the orbital angular momentum. In any spherically symmetric systems, energy eigenstates can be given by produce wave functions of the form (x) = R(r)Ylm (, ). (15)

In other words, we are parameterizing the position eigenbasis in terms of polar rather than Cartesian coordinates, x| = r, , |. (16)

The spherical harmonics are dened as the wave functions of angular momentum eigenstates Ylm (, ) = , |l, m . (17) Sakurai uses the notation n| and call them direction eigenkets. Clearly, the dentions of angular momemum eigenstates L2 |l, m Lz |l, m = l(l + 1) h2 |l, m , = mh |l, m , (18) (19)

translate to the dierential equations h 2 1 sin sin + 1 2 Ylm = l(l + 1) h2 Ylm , 2 2 sin h m Y = mh Ylm . i l (20) (21)

The latter equation is easy to solve: the azimuth dependence of the spherical harmonics must be eim . But guring out the polar angle dependence needs more work. The rest of the discussion here is on this issue. Both in the case of the harmonic oscillator and the Landau levels (energy levels of a charged particle in a uniform magnetic eld), it was useful to write
This corresponds to the separation of variables in HailtonJacobi theory, where the action S is written as a sum of dierent pieces S (r) + S (, ). Because the wave function corresponds to eiS/h , it is natural for the wave function to be given by a product of dierent pieces.

down an equation for the ground state of the form a|0 = 0. It was useful because it gave us a linear dierential equation instead of a quadratic one (e.g., Schr odinger equation). The linear dierential equations are far easier to solve. We can take the same strategy for the spherical harmonics.

2.1

Derivation of Spherical Harmonics

We know from the general representation theory of angular momenta that L |l, l = 0 because it cannot be lowered any more. Sakurai starts with |l, l instead. Of course you get the same result, but I nd this way somewhat less confusing. In the position representation, we nd 0 = , |L |l, l = h i e i cot Y l (, ) = 0. i l (22)

On the other hand, we know already that lh Yll = , |Lz |l, l = h l Y i l (23)

and hence the azimuth dependence of Yll is Yll (, ) = f ()eil . Therefore, Eq. (22) becomes d (24) i + il cot f () = 0. d This equation is solved easily, by writing it as 1 df = l cot d, f and integrating both sides to log f = l log sin + const. Therefore, f () = c sinl with an overall normalization constant c. Putting things together, we nd Yll (, ) = c sinl eil . 4 (28) (27) (26) (25)

The absolute value of c can be xed by the normalization 1= d|Yll |2 =

1 2

d cos
1 0

d|c|2 sin2l .

(29)

integral trivially gives a factor of 2 . The integral is most conveniently done using the variable x = cos , 1 = 2 |c|2
1 1

dx(1 x2 )l ..

(30)

Writing 1 x2 = (1 x)(1+ x), and further change the variable to x = 1+2t, 1 = 2 |c|2
0 1

2dt (2t(2 2t))l = 2 |c|2 22l+1

0

dt tl (1 t)l .

(31)

The integral is nothing but the Beta function

1

B (p, q ) =
0

dt tp1 (1 t)q1 =

(p)(q ) . (p + q )

(32)

Therefore, 1 = 2 |c|2 22l+1 We nd 1 (2l + 1)! . (34) 2l l! 4 The phase of c is xed by picking a convention. The commonly used convention (the same as Sakurais) is not to have an additional phase factor |c| = Yll (, ) = 1 2l l! (2l + 1)! sinl eil . 4 (35) (l + 1)(l + 1) l !l ! = 4 |c|2 22l . (2l + 2) (2l + 1)! (33)

Now that we know Yll , we can keep acting L+ on it to obtain all Ylm . Recall that the general discussion of angular momentum taught us that L+ |l, m = l(l + 1) m(m + 1)|l, m + 1 = (l m)(l + m + 1)|l, m + 1 . (36)

Then we nd Ylm (, ) = 1 1
l +m

h (l + m)(l m + 1) h (l + m 1)(l m + 2) 1 1 Yll (, )

h i e i cot h (2)(l + l 1) h (1)(l + l) i

l +m

(l m)! ei + i cot (2l)!(l + m)!

Yll (, )

(37)

Now the question is to bring it to a more usable form. The derivatives always give eigenvalues. But each time L+ acts, there is a factor of e+i and makes the eigenvalue of i/ increase one by one. Therefore, Ylm (, ) = 1 2l l! (2l + 1)! 4 (l m)! im d e (m 1) cot (2l)!(l + m)! d d + l cot sinl . d

d d (m 2) cot + (l 1) cot d d = 1 2l l! (2l + 1)(l m)! im d e (m 1) cot 4 (l + m)! d

d d (m 2) cot + (l 1) cot d d

d + l cot sinl . d (38)

Here we have cancelled factors of h and i. The next trick we need is to write d 1 d + k cot = sink d sink d Then Ylm (, ) = 1 2l l! 1 sin
(m2)

(39)

(2l + 1)(l m)! im 1 d e sin(m1) ( m 1) 4 (l + m)! d sin d 1 d sin(m2) sinl1 l 1 d d sin 6 1 d sinl sinl sinl d

1 = l 2 l! 1 = l 2 l! =

(2l + 1)(l m)! im m 1 d e sin 4 (l + m)! sin d (2l + 1)(l m)! im m d e sin 4 (l + m)! d cos

l +m

sin2l
l +m

sin2l (40)

(1)l+m 2l l!

dl+m (2l + 1)(l m)! im e (1 x2 )m/2 l+m (1 x2 )2l . 4 (l + m)! dx

In the last line, we used the variable x = cos .

2.2

Legendre Polynomials

Here are a few new notations. Legendre polynomials are dened by (1)l dl (1 x2 )l (41) 2l l! dxl dm Plm (x) = (1 x2 )m/2 m Pl (x). (42) dx The denition Eq. (42) works only for m > 0. Another way to write Plm is just putting them together, Pl (x) =
l +m (1)l 2 m/2 d (1 x ) (1 x2 )l . (43) 2l l! dxl+m This expression allows you to pick m < 0. Plm is called associated Legendre polynomials, and Plm (x) = Pl (x). The denitions look complicated, but they are just polynomials! Pl is a polynomial of order l. Plm has this funny factor (1 x2 )m/2 with a fractional power, but we will set x = cos in the end, and (1 x2 )m/2 = sinm . Remember is the polar angle, and we only consider 0 so that sin 0. Therefore, Plm is a polynomial of order m in sin and l m in cos . Now let us prove a surprising identity

Plm (x) =

Plm (x) = (1)m

l+m

(l m)! m P (x). (l + m)! l

(44)

d 2 l m Let us start with dx l+m (1 x ) for m > 0 which appears in Pl (x). We dlm 2 l expand it out and see how it can be related to dx lm (1 x )

dl+m dl+m 2 l (1 x ) = (1 x)l (1 + x)l dxl+m dxl+m 7

l +m

=
r =0

l+m Cr

dr (1 x)l dxr

dl+mr (1 + x)l . (45) dxl+mr

Even though the sum extends for 0 r l + m, the term in the rst parentheses vanishes for r > l while the second for l + m r > l. Therefore, the sum is taken only for m r l. Then dl+m (1 x2 )l l + m dx l! (1)r l! (1 x)lr (1 + x)rm = l+m Cr (l r)! (r m)! r =m
l m l

=
s=0

l+m Cm+s

(1)m+s l! l! (1 x)lms (1 + x)s . (l m s)! s!

(46)

In the last line, I rewrote it with r = m + s. Now I multiply it by (1 x2 )m and divide by it, dl+m (1 x2 )l dxl+m 1 = (1 x2 )m = = = 1 (1 x2 )m 1 (1 x2 )m

l m l+m Cs+s s=0 l m s=0 l m s=0

(1)m+s l! l! (1 x)ls (1 + x)m+s (l m s)! s!

(l + m)! (1)m+s l! l! (1 x)ls (1 + x)m+s (m + s)!(l s)! (l m s)! s! (l + m)! (1)m+s l! l! (1 x)ls (1 + x)m+s s!(l m s)! (l s)! (m + s)!
l m s=0

(l + m)! 1 (l m)! (1 x2 )m
s

(l m)! (1)m+s s!(l m s)! dlms (1 + x)m+s dxlms ds (1 x)ls s dx dlms (1 + x)m+s l m s dx (47)
(1)l (1 2l l !

(1)s

d (1 x)ls dxs
l m lm Cs s=0 l m

(l + m)! (1)m = (l m)! (1 x2 )m =

(l + m)! (1)m d (1 x2 )l . (l m)! (1 x2 )m dxlm x2 )m/2 , we nd

Multiplying both sides of the equation by Plm (x) =

l +m (1)l 2 m/2 d (1 x ) (1 x2 )l 2l l! dxl+m

(1)l (l + m)! dlm 1 (1)m l (1 x2 )l (l m)! 2 l! (1 x2 )m/2 dxlm (l + m)! = (1)m Plm (x). (l m)!

(48)

This is indeed Eq. (44). The orthogonality relation among Legendre polynomials is important:
1 1 m dxPn (x)Plm (x) =

2 (l + m)! n,l . 2l + 1 (l m)!

(49)

Note that both polynomials share the same m. It can be shown as follows. In this discussion, we assume m 0. But m 0 case follows from Eq. (44). We substitute in the explicit expression,
1 1 m dxPn (x)Plm (x) 1

n+m (1)n 2 m/2 d (1 x ) (1 x2 )n n n + m 2 n! dx 1 l l +m (1) d (1 x2 )m/2 l+m (1 x2 )l l 2 l! dx n+l 1 (1) dn+m 2 m = n+l dx(1 x ) (1 x2 )n n + m 2 n !l ! 1 dx

dx

dl+m (1 x2 )l .(50) l + m dx

To show the orthogonality when n = l, let us assume n > l without a loss of generality. The point is to keep doing integration by parts so that dn+m /dxn+m 4 acting on (1 x2 )n+m acts on the rest of the integrand. But the rest is a polynomial of order 2m + 2l (l + m) = l + m < n + m, and its (n + m)-th derivative vanishes. At each integration by parts, the surface term also vanishes. For the rst m-times, the surfact term vanishes because of (1 x2 )m factor. For the remaining n-times, it vanishes because of the (1 x2 )n factor. This proves the orthogonality. When n = l, we again go through the same procedure and only the top power in x contributes,
1 1

dxPlm (x)Plm (x) (1)l+l (1)l+m 2l+l l!l!

1 1

dx(1 x2 )l

l +m dl+m 2 m d (1 x ) (1 x2 )l dxl+m dxl+m

l +m 1 (1)l+l dl+m l +m 2 l m 2 m d ( 1) dx (1 x ) ( 1) ( x ) (1)l (x2 )l l + m l + m 2l+l l!l! dx dx 1 1 dl+m 2 m (2l)! lm 1 dx(1 x2 )l x (x ) = 2l 2 l !l ! 1 dxl+m (l m)! 1 1 (2l)! = 2l dx(1 x2 )l (l + m)! . (51) 2 l !l ! 1 (l m)!

Change the variable to x = 1 + 2t,

1 1

dxPlm (x)Plm (x) = = = = =

1 (2l)!(l + m)! 1 2dt(2t(2 2t))l 2 l 2 l!l! (l m)! 0 1 2 (2l)!(l + m)! dttl (1 t)l l!l! (l m)! 0 2 (2l)!(l + m)! (l + 1)(l + 1) l!l! (l m)! (2l + 2) 2 (2l)!(l + m)! (2l + 1)! (l m)! 2 (l + m)! 2l + 1 (l m)!

(52)

2.3

More Conventional Expression

(2l + 1)(l m)! im m e Pl (cos ). 4 (l + m)!

Compared to Eq. (40), we see that Plm gives Ylm , Ylm = (1)m A special case of this is (2l + 1) Pl (cos ). (54) 4 When m < 0, an alternative expression is obtained by using the identity Eq. (44), Yl 0 = Yl m = (2l + 1)(l |m|)! im |m| e Pl (cos ). 4 (l + |m|)! Ylm = (1)m (Ylm ) . 10 (55) (53)

In particular, this expression shows a relation (56)

2.4

Orthonormality and Completeness

You can verify the orthonormality of spherical harmonics explicitly. Here and below, refers to the polar coordinates , , and the integration volume is d = d cos d. dYlm ()(Ylm ()) = (1)m+m
1

(2l + 1)(l m)! 4 (l + m)!

2 0

(2l + 1)(l m )! 4 (l + m )! (57)

d cos
1

dPlm (cos )Plm (cos )eim eim .

Because integral vanishes unless m = m , = 2m,m = 2m,m (2l + 1)(l m)! 4 (l + m)! (2l + 1)(l m)! 4 (l + m)! (2l + 1)(l m)! 4 (l + m)!
1 1

dxPlm (x)Plm (x)

(2l + 1)(l m)! 2 (l + m)! l,l 4 (l + m)! 2l + 1 (l m)!

= 2m,m l,l = m,m l,l .

(2l + 1)(l m)! 2 (l + m)! l,l 4 (l + m)! 2l + 1 (l m)! (58)

This is an explicit verication of the expected orthonormality m,m l,l = l , m |l, m = d l , m |, , |l, m = d(Ylm ) Ylm . (59)

The completeness relation is 2 ( ) = = =

l,m

, | , |l, m l, m|
l,m

Ylm (, )(Ylm ( , )) .

(60)

Here, 2 ( ) = (cos cos ) ( ) = 11 ( ) ( ). sin (61)

2.5

Examples
(62) (63) (64) (65) (66)

It is useful to look at a few examples: 1 Y00 = , 4 3 Y11 = sin ei , 8 Y10 = Y22 = Y21 = Y20 = 3 cos , 4 15 sin2 e2i , 32 15 sin cos ei , 8

5 (3 cos2 1). (67) 16 In spectroscopic symbols, l = 0, 1, 2, 3, 4, correspond to s, p, d, f, g, orbitals. In chemistry and solid-state physics, you see symbols like px , dx2 y2 . They refer to certain linear combinations of spherical harmonics. The general rule is to rst multiply the spherical harmonics by rl , and you nd 1 (68) Y00 = , 4 3 rY11 = (x iy ), (69) 8 rY10 = r2 Y22 = r2 Y21 = r2 Y20 =

3 z, 4 15 (x iy )2 , 32 15 (x iy )z, 8 5 (3z 2 r2 ). 16

(70) (71) (72) (73)

s for sharp, p for principal, d for diuse, f for fundamental, and the rest is just alphabetical.

12

Then you look at x, y , z dependences to identify a particular orbital. Note that the dependence on r should not be taken seriously; it is supposed to be multiplied by a radial wave function anyway. We multiplied spherical harmonics by rl just to make to expressions become polynomials in x, y , z . From these expressions, it is clear that the pz orbital corresponds to Y10 , + Y11 )/ 2 and py to ( while px to (Y11 Y11 Y11 )/i 2. dx2 y2 corresponds 1 2 to (Y22 + Y2 )/ 2, dyz to (Y21 Y2 )/i 2. Im sure you have seen shapes of spherical harmonics in textbooks. It is important to understand what they actually are. In many cases, what is shown is a surface given by points r = |Ylm (, )|2 . (74)

In other words, the distance of the surface from the origin along a direction is determined by the probability of nding the particle along that direction. They actually do not represent the shapes of the wave function. They just show along which direction the probability is big or small. In certain cases, though, these plots do represent the shapes of the actual wave function. Remember that the actual wave function has the radial wave function on top of the spherical harmonics. Suppose the radial wave function is a smoothly decaying function, say, er/a0 . Now you try to draw a surface of constant probability density in three dimensions. Then along the directions where |Ylm |2 is larger, the constant probability is attained even at higher r; but along the directions with small |Ylm |2 , you need to go closer to the origin to get the same probability density. Then the plot mentioned above can approximate the shape of the actual wave function. But if you want to interpret the plots in this manner, it obviously depends on the details of the radial wave function, and what value you chose for the surface of constant probability density. Just presenting the shapes of, say, pz orbitals independent of n (the principal quantum number) is therefore misleading.

2.6

Mathematica

It is useful to know that Mathematica has built-in commands for the spherical harmonics SphericalHarmonicY[l,m,,] for Ylm (, ), the Legendre polynomials LegendreP[n,x] for Pn (x), and the associated Legendre polym nomials LegendreP[n,m,x] for Pn (x).

13