You are on page 1of 149

8.

06: Quantum Mechanics III

Tony Zhang

Spring 2017

These are my lecture notes from 8.06, Quantum Mechanics III, at the
Massachusetts Institute of Technology, taught this semester (Spring 2017) by
Professor Barton Zwiebach1 . I attended recitation sections by Professor Maxim
Metlitski2 .
I wrote these lecture notes in LATEX in real time during lectures, so there
may be errors and typos. I have shamelessly pillaged Evan Chen’s formatting
commands. Should you encounter an error in the notes, wish to suggest im-
provements, or alert me to a failure on my part to keep the web notes updated,
please contact me at txz@mit.edu.
This document was last modified 2017-05-29. The permalink to these notes
is http://web.mit.edu/txz/www/links.html.

1 zwiebach@mit.edu
2 mmetlits@mit.edu

i
Tony Zhang Contents

Contents

I Time-independent perturbation theory 1

1 February 7, 2017 2
1.1 Administrivia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Time-independent Perturbation Theory . . . . . . . . . . . . . . 2
1.3 Validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Feburary 8, 2017 (recitation) 5


2.1 Example: Shifted harmonic oscillator . . . . . . . . . . . . . . . . 5
2.2 Example: van der Waals forces . . . . . . . . . . . . . . . . . . . 5

3 February 14, 2017 6


3.1 More nondegenerate perturbation theory . . . . . . . . . . . . . . 6
4
3.1.1 Example: x anharmonic oscillator . . . . . . . . . . . . . 6
3.2 Degenerate perturbation theory . . . . . . . . . . . . . . . . . . . 7

4 February 16, 2017 9


4.1 First-order state shift . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 The general case . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

5 February 21, 2017 (recitation) 12


5.1 Two-dimensional anharmonic oscillator . . . . . . . . . . . . . . . 12

II The hydrogen atom 14

6 February 22, 2017 (recitation) 15


6.1 Dirac equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

7 February 23, 2016 17


7.1 Hydrogen review . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
7.2 Spin-orbit coupling . . . . . . . . . . . . . . . . . . . . . . . . . . 17
7.2.1 Pauli equation . . . . . . . . . . . . . . . . . . . . . . . . 18
7.3 Dirac equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

8 February 27, 2017 (recitation) 21

9 February 28, 2017 23


9.1 The Darwin term . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
9.2 The relativistic correction . . . . . . . . . . . . . . . . . . . . . . 24
9.3 Spin-orbit coupling . . . . . . . . . . . . . . . . . . . . . . . . . . 25
9.4 Remarks on Zeeman . . . . . . . . . . . . . . . . . . . . . . . . . 26

ii
Tony Zhang Contents

10 March 1, 2017 (recitation) 27


10.1 Fine structure from Dirac . . . . . . . . . . . . . . . . . . . . . . 27

III The WKB approximation 29

11 March 2, 2017 30
11.1 WKB; the semiclassical approximation . . . . . . . . . . . . . . . 30
11.2 Local momentum, de Broglie wavelength . . . . . . . . . . . . . . 31
11.3 Solving the equation . . . . . . . . . . . . . . . . . . . . . . . . . 31
11.4 Validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
11.5 The connection formula . . . . . . . . . . . . . . . . . . . . . . . 33

12 March 6, 2017 (recitation) 35


12.1 Gamow’s theory of α-decay . . . . . . . . . . . . . . . . . . . . . 35

13 March 7, 2017 36
13.1 Establishing the connection formulae . . . . . . . . . . . . . . . . 36
13.2 Arrows and asymmetric validity . . . . . . . . . . . . . . . . . . . 39
13.3 Tunneling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

14 March 8, 2017 (recitation) 40


14.1 Tunneling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

IV Time-dependent perturbation theory 41

15 March 9, 2017 42
15.1 Time-dependent perturbation theory . . . . . . . . . . . . . . . . 42
15.2 The interaction picture . . . . . . . . . . . . . . . . . . . . . . . . 42
15.3 Exact evolution in the interaction picture . . . . . . . . . . . . . 43
15.3.1 Example: two-state system . . . . . . . . . . . . . . . . . 43
15.4 Perturbative solution . . . . . . . . . . . . . . . . . . . . . . . . . 45

16 March 13, 2017 (recitation) 46


16.1 Instantaneous approximation . . . . . . . . . . . . . . . . . . . . 46
16.1.1 Example: β-decay . . . . . . . . . . . . . . . . . . . . . . 46

17 March 15, 2017 (recitation-cum-lecture) 48


17.1 Time-dependent perturbation theory: review . . . . . . . . . . . 48
17.2 Transition probability . . . . . . . . . . . . . . . . . . . . . . . . 48
17.3 First-order coefficients . . . . . . . . . . . . . . . . . . . . . . . . 49
17.4 Fermi’s Golden Rule with constant transitions: intro . . . . . . . 49
17.4.1 Example: Auto-ionization of helium . . . . . . . . . . . . 50

iii
Tony Zhang Contents

18 March 16, 2017 51


18.1 Fermi’s Golden Rule with constant transitions . . . . . . . . . . . 51
18.2 Harmonic perturbations: introduction . . . . . . . . . . . . . . . 53

19 March 21, 2017 55


19.1 Fermi’s Golden Rule with harmonic perturbation . . . . . . . . . 55
19.2 Example: Ionization of hydrogen . . . . . . . . . . . . . . . . . . 56
19.3 Light and atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

20 March 22, 2017 (recitation) 59


20.1 Scattering and the Born approximation . . . . . . . . . . . . . . 59
20.2 About probabilities . . . . . . . . . . . . . . . . . . . . . . . . . . 60

21 Apri 4, 2017 61
21.1 Einstein’s argument for spontaneous emission . . . . . . . . . . . 61
21.2 Atom-light interaction . . . . . . . . . . . . . . . . . . . . . . . . 62

22 April 5, 2017 (recitation) 65


22.1 Fermi’s Golden Rule TODO TODO TDOO TODOT ODO TODO 65
22.2 Selection rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

V The adiabatic approximation 67

23 April 6, 2017 68
23.1 Classical adiabatic change . . . . . . . . . . . . . . . . . . . . . . 68
23.2 Quantum mechanical motivation . . . . . . . . . . . . . . . . . . 69
23.3 Quantum adiabatic evolution . . . . . . . . . . . . . . . . . . . . 70

24 April 11, 2017 72

25 April 12, 2017 (recitation) TODO TODO TODO TODO 73

26 April 14, 2017 75


26.1 Landau-Zener transition probabilities . . . . . . . . . . . . . . . . 75
26.1.1 The adiabatic timescale . . . . . . . . . . . . . . . . . . . 75
26.1.2 Transition probabilities . . . . . . . . . . . . . . . . . . . 76
26.2 The Berry phase . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

27 April 20, 2017 79


27.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
27.2 More on Berry’s phase . . . . . . . . . . . . . . . . . . . . . . . . 79
27.3 Molecules and the Born-Oppenheimer approximation . . . . . . . 80

iv
Tony Zhang Contents

28 April 24, 2017 (recitation) 83


28.1 More about the Berry phase . . . . . . . . . . . . . . . . . . . . . 83
28.1.1 Example: spin- 12 in B-field . . . . . . . . . . . . . . . . . 83

29 April 25, 2017 85


29.1 Born-Oppenheimer: the nuclear wavefunction . . . . . . . . . . . 85
29.2 Example: the H2 + ion . . . . . . . . . . . . . . . . . . . . . . . . 88

VI Scattering 89

30 April 26, 2017 (recitation) 90


30.1 Scattering warm-up . . . . . . . . . . . . . . . . . . . . . . . . . 90

31 April 27, 2017 93


31.1 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
31.2 Partial waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
31.3 Phase shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

32 May 1, 2017 (recitation) 97


32.1 The optical theorem . . . . . . . . . . . . . . . . . . . . . . . . . 97

33 May 2, 2017 99
33.1 More about phase shifts . . . . . . . . . . . . . . . . . . . . . . . 99
33.2 Example: hard sphere scattering . . . . . . . . . . . . . . . . . . 101
33.3 General computation of the phase shift . . . . . . . . . . . . . . . 103
33.4 Intuition for phase shifts . . . . . . . . . . . . . . . . . . . . . . . 103
33.4.1 Impact parameters . . . . . . . . . . . . . . . . . . . . . . 103
33.4.2 Partial wave unitarity . . . . . . . . . . . . . . . . . . . . 104

34 May 3, 2017 (recitation) 105


34.1 Wavepacket scattering . . . . . . . . . . . . . . . . . . . . . . . . 105

35 May 4, 2017 107


35.1 Scattering integral equation . . . . . . . . . . . . . . . . . . . . . 107
35.2 The Born approximation . . . . . . . . . . . . . . . . . . . . . . . 109
35.2.1 Radial potentials . . . . . . . . . . . . . . . . . . . . . . . 110

36 May 8, 2017 (recitation) 111


36.1 Wavepacket scattering cross-section . . . . . . . . . . . . . . . . . 111
36.2 A word on the Born aproximation . . . . . . . . . . . . . . . . . 112

VII Identical particles 114


v
Tony Zhang Contents

37 May 9, 2017 115


37.1 Identical particles . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
37.1.1 Identical particles in classical mechanics . . . . . . . . . . 115

38 May 10, 2017 (recitation) 118


38.1 More on the Born approximation . . . . . . . . . . . . . . . . . . 118

39 May 11, 2017 119


39.1 The symmetric group . . . . . . . . . . . . . . . . . . . . . . . . 119
39.2 Symmetrizers and antisymmetrizers . . . . . . . . . . . . . . . . 120

40 May 15, 2017 (recitation) 122


40.1 About identical particles . . . . . . . . . . . . . . . . . . . . . . . 122

41 May 16, 2017 124


41.1 The symmetrization postulate . . . . . . . . . . . . . . . . . . . . 124
41.2 Building antisymmetric states . . . . . . . . . . . . . . . . . . . . 124
41.3 Occupation numbers . . . . . . . . . . . . . . . . . . . . . . . . . 125
41.4 Particles in product spaces . . . . . . . . . . . . . . . . . . . . . 126

42 Exchange forces (home reading) 127

43 May 17, 2017 (recitation) 127

44 May 18, 2017 128


44.1 Identical composite systems . . . . . . . . . . . . . . . . . . . . . 128
44.2 Counting states and distributions . . . . . . . . . . . . . . . . . . 129
44.2.1 Maxwell-Boltzmann statistics (distinguishable particles) . 129
44.2.2 Fermi-Dirac statistics (identical fermions) . . . . . . . . . 129
44.2.3 Bose-Einstein statistics (identical bosons) . . . . . . . . . 129
44.2.4 Deriving distributions . . . . . . . . . . . . . . . . . . . . 129

A Useful formulae 131


A.1 Wave mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
A.2 Spin- 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
A.3 Harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 131
A.4 Angular momentum . . . . . . . . . . . . . . . . . . . . . . . . . 132
A.5 Hydrogen atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
A.5.1 Gross structure . . . . . . . . . . . . . . . . . . . . . . . . 132
A.5.2 Fine structure . . . . . . . . . . . . . . . . . . . . . . . . . 133
A.5.3 Zeeman effect . . . . . . . . . . . . . . . . . . . . . . . . . 133
A.6 Time-independent perturbation theory . . . . . . . . . . . . . . . 133

vi
Tony Zhang Contents

A.6.1 Nondegenerate case . . . . . . . . . . . . . . . . . . . . . 133


A.6.2 Degenerate case . . . . . . . . . . . . . . . . . . . . . . . . 134
A.7 WKB approximation . . . . . . . . . . . . . . . . . . . . . . . . . 134
A.8 Time-dependent perturbation theory . . . . . . . . . . . . . . . . 135
A.8.1 Fermi’s Golden Rule . . . . . . . . . . . . . . . . . . . . . 136
A.8.2 Emission rates . . . . . . . . . . . . . . . . . . . . . . . . 136
A.8.3 Selection rules for hydrogen . . . . . . . . . . . . . . . . . 136
A.9 Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
A.10 Adiabatic approximation . . . . . . . . . . . . . . . . . . . . . . . 137
A.10.1 Landau-Zener transitions . . . . . . . . . . . . . . . . . . 137
A.10.2 Berry’s phase . . . . . . . . . . . . . . . . . . . . . . . . . 137
A.10.3 Molecules and Born-Oppenheimer . . . . . . . . . . . . . 137
A.11 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
A.11.1 Partial waves and phase shifts . . . . . . . . . . . . . . . . 138
A.11.2 Integral scattering and the Born approximation . . . . . . 139
A.12 Identical particles . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

B Useful math 141


B.1 Trigonometric integrals . . . . . . . . . . . . . . . . . . . . . . . . 141
B.2 Gaussian integrals . . . . . . . . . . . . . . . . . . . . . . . . . . 141
B.3 Gamma function . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
B.4 Spherical harmonics . . . . . . . . . . . . . . . . . . . . . . . . . 141
B.5 Generalized Laguerre polynomials . . . . . . . . . . . . . . . . . 142
B.6 Spherical Bessel functions . . . . . . . . . . . . . . . . . . . . . . 142

vii
Tony Zhang

Part I

Time-independent perturbation
theory

1
Tony Zhang 1 February 7, 2017

1 February 7, 2017
1.1 Administrivia
• There’s stuff on Stellar
• There will be an in-class midterm on Thursday, March 23, 2017.
• Weekly psets, due Tuesdays at 5pm.

• A term paper exists. See Stellar for more. Don’t worry about it until the
class gets busy (April).
• The structure of these lectures will follow that of Aram Harrow’s lecture
notes.

• This course is different from 8.04 or 8.05... “we do things that are more
practical here”. A lot of what we do in this course has to do with
approximation.

1.2 Time-independent Perturbation Theory


Before beginning, let’s note that perturbation theory is super useful:

• hyperfine splitting

• spin-orbit coupling
• van der Waals forces

Today we’ll consider the nondegenerate case. We’ll see that it’s the simpler
case.
Let’s say we have a Hamiltonian H that we’d like to study. It turns out that
H = H 0 + δH, where we understand the original Hamiltonian H 0 really well,
and where δH is “small”.
What does it mean to be “small”? Does it mean it has small matrix elements?
Not quite.
Instead, let’s write
H(λ) = H 0 + λ δH (1.1)
where λ is a small, dimensionless number.
We assume that everything about H 0 is known. That is, we know its
spectrum
H 0 |n(0) i = En(0) |n(0) i (1.2)
for n ∈ Z. (In principle, we could have uncountable spectra too.)
(0) (0)
We impose non-degeneracy: that is, En 6= Em for n 6= m. Non-degeneracy
lets us characterize (label, “track”) our eigenstates by their eigenvalues.
We want to find the spectrum of H(λ):

H(λ) |niλ = En (λ) |niλ . (1.3)

2
Tony Zhang 1 February 7, 2017

If λ = 0, we should recover the original spectrum. Otherwise, we make the


assumption that we can write the new eigenvalues and eigenvectors as a power
series:

X
|niλ = λk |n(k) i (1.4)
k=0
X∞
En (λ) = λk En(k) (1.5)
k=0

(k)
In the spirit of power series, the |n(k) i and En , the unknown “coefficients”
on λ, are λ independent.
Formally,
1 ∂ k |niλ

(k)
|n i= (1.6)
k! ∂λk λ=0
(k)
and similarly for En .
Write the eigenvalue equation with these new eigenvalues and eigenvectors
as a power series in λ:
0 = (H 0 + λ δH − En (λ)) |niλ
= ((H 0 − En(0) ) − λ(En(1) − δH) − λ2 En(2) − . . . )
(|n(0) i + λ |n(1) i + λ2 |n(2) i + . . . )
Joke. “The difficulty with perturbation theory is the labels.”

The polynomial on the right hand side must vanish identically, so its coeffi-
cients all vanish:
λ0 (H 0 − En(0) ) |n(0) i = 0 (1.7)
1 0
λ (H − En(0) ) |n(1) i = (En(1) − δH) |n (0)
i (1.8)
λ2 (H 0 − En(0) ) |n(2) i = (En(1) − δH) |n(1) i + En(2) |n(0) i (1.9)
.. ..
. .
λk (H 0 − En(0) ) |n(k) i = (En(1) − δH) |n(k−1) i + En(2) |n(k−2) i + · · · + En(k) |n(0) i
(1.10)
These equations can be solved recursively.
Does |n(1) i have to be orthogonal to |n(0) i? Note that Equation 1.9, which
defines it, holds if we add arbitrary |n(0) i components on |n(1) i, by Equation 1.7.
Indeed, we can and will, without loss of generality, assume that hn(0) |n(k) i = 0
for k 6= 0.
Let’s not worry about finding the eigenvector corrections first. We’ll look for
(k)
the En .
Adjoining a hn(0) | to Equation 1.8 gives us

En(1) = hn(0) |δH|n(0) i . (1.11)

Some people say this is the most important practical result in quantum mechanics.
Doing the same to Equation 1.9 gives
En(2) = hn(0) |δH|n(1) i .

3
Tony Zhang 1 February 7, 2017

In general, we get
En(k) = hn(0) |δH|n(k−1) i (1.12)

OK, great, but these all require the eigenvector corrections |n(k) i. So let’s
compute those.
We resolve the identity over the spectrum of H 0 :
X
1= |k (0) i hk (0) | .
k6=n

Then we can stick it into Equation 1.8 to get

(H 0 − En(0) ) |n(1) i = (En(1) − δH) |n(0) i


X
= |k (0) i hk (0) | (En(1) − δH) |n(0) i
k6=n
X
=− |k (0) i hk (0) |δH|n(0) i
| {z }
k6=n δHkn
(0)
X |k i
|n(1) i = − (0) (0)
δHkn
k6=n Ek − En

We can thus find the second-order energy shift:

En(2) = hn(0) |δH|n(1) i (1.13)


X hn(0) |δH|k (0) i δHkn
= (0) (0)
(1.14)
k6=n En − Ek
X |δHnk |2
= (0) (0)
(1.15)
k6=n En − Ek

1.3 Validity
How valid is our perturbative solution? Suppose we have H (0) = diag(E1 , E2 )
with  
(0) E1 λV
H + λV̂ =
λV ∗ E2

We can compute the spectrum analytically as


s
E1 + E2 E1 − E2 λ2 |V |2
E± (λ) = ± 1+  .
2 2 E1 −E2 2
2

The Taylor expansion of the square root gives some series with finite radius of
convergence (it’s 1, actually). So a perturbative expansion is only valid when

E1 − E2
λ|V |<
.
2

We see a hint of the difficulties to come when the original spectrum becomes
degenerate.

4
Tony Zhang 2 Feburary 8, 2017 (recitation)

2 Feburary 8, 2017 (recitation)


2.1 Example: Shifted harmonic oscillator
Perturb the harmonic oscillator Hamiltonian as:
p2 1
H= + mω 2 x2 − qEx. (2.1)
2m 2
Completing the square in the x terms actually shows that the new wavefunctions
are displaced versions of the old.
Using perturbation theory, we can find:

• the first-order energy corrections all vanish


• the second order energy corrections are simple
• the first-order wavefunction shifts agree with the answer that the new
wavefunctions are just displaced versions of the old

2.2 Example: van der Waals forces


We’ll show that two distant interacting hydrogen atoms to so under the van der
Waals potential V (R) = −CR−6 for C > 0. By distant, we mean that R  a0 ,
where a0 is the Bohr radius.
Model two interacting hydrogen atoms by the following Hamiltonian:

p2A e2 p2 e2
H (0) = − + B − (2.2)
2me 4πε0 |~rA | 2me 4πε0 |~rB |
where pi are the electron momenta and ~ri are the electron displacements from
their respective nuclei. The energies are just
 
1 1
En,n0 = −13.6 eV +
n2 (n0 )2

But we’ve neglected electron-electron, elctron-other-nucleus, and nucleus-


nucleus interactions. So we add a perturbation
!
e2 1 1 1 1
δH = − + − − (2.3)
4πε0 |R ~ − ~rA | |R
~ + ~rB | |R~ + ~rB − ~rA | |R|
~
e2
= (~rA · ~rB − 3(~rA · R̂)(~rB · R̂)) + O(R−4 ) (2.4)
4πε0 R3

Now assume both electrons are in their ground states.


Joke. The first-order correction vanishes... so we will need to work harder.
[Learns class is done in two minutes.] ... so we will not work much harder.

(2) e2 a60
δE(1,0,0),(1,0,0) ∼ −
4πε0 a0 R6

5
Tony Zhang 3 February 14, 2017

3 February 14, 2017


Recall the following from nondegenerate perturbation theory.
H(λ) = H (0) + λ δH (3.1)
X δHkn
|niλ = |n(0) i − λ (0) (0)
+ O(λ2 ) (3.2)
k6=n Ek − En
X |δHnk |2
En (λ) = En(0) + λ δHnn − λ2 (0) (0)
+ O(λ3 ) (3.3)
k6=n Ek − En

Joke. *Loud banging on door* Logs Valentine’s serenade happens!

3.1 More nondegenerate perturbation theory


Why is the degenerate case (i.e. with degeneracies in H (0) ) harder? In short: it
becomes harder to track states when there are degeneracies.
Before continuing, we make some remarks.

1. The first order corected ground state energy overstates the true ground
state energy. By the Variational principle:
(0)
E0 + λ δH00 = h0(0) |(H (0) + λ δH)|0(0) i
= h0(0) |H(λ)|0(0) i
≥ E0 (λ)

2. The second-order correction to the energies exhibits level repulsion. Write


this correction as
X |δHnk |2 X |δHnk |2
−λ2 (0) (0)
+ λ2 (0) (0)
k>n Ek − En k<n En − Ek

whence higher-energy states push this correction down, and lower-energy


states push this correction up. This effect explains why level crossings are
“rare” (intuitively, it requires some “effort” to get levels to cross).

3.1.1 Example: x4 anharmonic oscillator

Consider the harmonic oscillator with an x4 perturbation:


p̂2 1
H (0) = + mω 2 x̂2
2m 2
m2 ω 3 4
H(λ) = H (0) + λ x̂

The first order ground state energy correction is then:
(1) m2 ω 3
E0 = δH00 = h0|x̂4 |0i

m2 ω 3 h̄2
= h0|(â + ↠)4 |0i
h̄ 4m2 ω 2
1
= h̄ω h0|(â + ↠)4 |0i
4 | {z }
3

6
Tony Zhang 3 February 14, 2017

Remark 3.1. To compute expressions like h0|(â + ↠)4 |0i, “peel off” one term
from the operator power at a time; don’t expand the whole thing. Write it like
(â + ↠)(â + ↠)3 .

OK, so we have the first order ground state energy correction:


 
(0) (1) 1 3
E0 + λE0 = h̄ω 1 + λ .
2 2

To get the second-order correction, we have

(2)
X |δH0k |2
E0 =− (0) (0)
.
k>0 Ek − E0

Now we want the values δH0k = 14 h̄ω h0|(a + a† )4 |ki. We make some notes:

• The braket is vanishes unless k ≤ 4.


• The braket vanishes unless k is even, by parity reasons.

Therefore, we only need compute δH02 and δH04 . These turn out to be:

3 24
δH02 = √ h̄ω δH04 = h̄ω
2 4
whence
(2) 21
E0 =− h̄ω.
8
Then our second-order result is
 
(0) (1) 2 (2) 1 3 21 2 3
E0 + λE0 +λ E0 = h̄ω 1 + λ − λ + O(λ )
2 2 4

People have computed this series to super-high order. It turns out that this
series fails to converge for any λ 6= 0. It’s an asymptotic expansion. While
the series doesn’t converge, it does give a good approximation if you terminate
summation early. In practice, you add to your partial sum until terms start
growing.
Joke. *I get Muses-serenaded* “I guess 8.06 students are very popular; maybe
we’ll get a third!” — Zwiebach, paraphrased

3.2 Degenerate perturbation theory


Consider the following:
   
1 0 0 1
H(λ) = +λ .
0 1 1 0

Naively, we might take the standard basis as our eigenbasis, and compute
corrections 1 + λδHii = 1. But the actual eigenvalues are 1 ± λ, with eigenvectors
√1 (1, ±1)T .
2
For arbitrarily small λ > 0, the eigenvectors change discontinuously. But
was there actually discontinuous behavior? Not really. Our original eigenvectors
actually lived in a two-dimensional subspace of eigenvectors.

7
Tony Zhang 3 February 14, 2017

Now let’s develop the general theory. So suppose now that H (0) has spectrum
(0)
in which energy En has degeneracy N . Call this eigenspace Vn , and let the
other eigenspaces’ span be V̂ , whence our total state space is H = Vn ⊕ V̂ .
Restrict attention to Vn , and take some orthonormal basis |n(0) , ki for
k = 1, . . . , N . Let |p(0) i be a basis for V̂ .
Again, assume states and energies vary continuously as

|n; kiλ = |n(0) , ki + λ |n(1) ; ki + λ2 |n(2) ; ki + O(λ3 )


(1) (2)
En,k (λ) = En(0) + λEn,k + λ2 En,k + O(λ3 )

As before, assume without loss of generality that the corrections |n(j) ; ki are
orthogonal to |n(0) ; ki. They still may have components in the orthogonal
space V̂ , however.
Now using the eigenvalue equation, we have

(H (0) − En(0) ) |n(0) ; ki = 0 (3.4)


(0) (1)
(H − En(0) ) |n(1) ; ki = (En,k − δH) |n (0)
; ki (3.5)
(1) (2)
(H (0) − En(0) ) |n(2) ; ki = (En,k − δH) |n(1) ; ki + En,k |n(0) ; ki (3.6)

as before.
We now assume that δH completely breaks the degeneracy to first order in
λ. Let’s see what happens when we bra Equation 3.5 with hn(0) ; l|:
(1)
0 = hn(0) , l|(En,k − δH)|n(0) , ki
(1)
hn(0) ; l|δH|n(0) , ki = En,k δlk
(1)
δHnl,nk = En,k δlk

So we must have the Vn block of δH be diagonal. Assuming that is so, setting


l = k gives the first-order energy correction,
(1)
En,k = (δH)nk,nk (3.7)

8
Tony Zhang 4 February 16, 2017

4 February 16, 2017


Recall that we have

(H (0) − En(0) ) |n(0) ; ki = 0 (4.1)


(0) (1)
(H − En(0) ) |n(1) ; ki = (En,k − δH) |n (0)
; ki (4.2)
(1) (2)
(H (0) − En(0) ) |n(2) ; ki = (En,k − δH) |n(1) ; ki + En,k |n(0) ; ki (4.3)

Further, recall that for states originally in a degenerate subspace V to be


trackable at all (that is, for them to look like power series when we vary λ), we
need δH to be diagonal in V.
We make some remarks:

1. Equation 3.7 is always true, even if the perturbation δH does not break
the degeneracy to first order.
2. For the results of degenerate perturbation theory to hold, δH must have
diagonal V block.
3. δH is diagonal in V if we can find some Hermitian K with [δH, K] = 0
(over our entire state space H) such that the basis vectors we chose in V
are eigenstates of K with distinct eigenvalues.

Proof. Assume k 6= l. Then

0 = hn(0) ; k|[δH, K]|n(0) ; li = (λl − λk ) hn(0) ; k|δH|n(0) ; li .


| {z }
6=0

4.1 First-order state shift


Now we want to compute |n(1) ; ki. Use a resolution of the identity on Equa-
tion 4.2, which we note only gives |n(1) ; ki modulo V.
(1)
X
(H (0) − En(0) ) |n(1) ; ki = |p(0) i hp(0) | (En,k − δH) |n(0) ; ki
p
X
=− |p(0) i δHp,nk
p
X |p(0) i
|n(1) ; ki = (0) (0)
δHp,nk + |ψn,k i (4.4)
p En − Ep | {z }
| {z } ∈Vn
∈V̂

where we add the |ψn,k i (like +C in integration). It turns out that we get
inconsistencies if we do not include this term.

9
Tony Zhang 4 February 16, 2017

We don’t want to find |n(2) ; ki at this point. But let’s bra Equation 4.3 with
(0)
hn ; l| and see what we get:
(1) (2)
0 = hn(0) ; l|(En,k − δH)|n(1) ; ki + hn(0) ; l|En,k |n(0) ; ki
(1)
X δHp,nk |p(0) i
= hn(0) ; l| (En,k − δH) (0) (0)
p En − Ep
(1)
+ hn(0) ; l|(En,k − δH)|ψn,k i
(2)
+ En,k δkl
X δHp,nk δHnl,p (1) (1) (2)
0=− (0) (0)
+ (En,k − En,l ) hn(0) ; l|ψn,k i + Enk δkl (4.5)
p E n − E p

We now focus on Equation 4.5. Setting l = k, we get

(2)
X δHnk,p δHp,nk
Enk = (0) (0)
(4.6)
p En − Ep

which gives the correct expression for the second-order energy shift.
Finally, to figure out the wavefunction correction, take k 6= l in Equation 4.5.
Then
X δHp,nk δHnl,p (1) (1)
− (0) (0)
+ (En,k − En,l ) hn(0) ; l|ψn,k i = 0
p E n − E p

which yields
1 X δHp,nk δHnl,p
hn(0) ; l|ψn,k i = (1) (1) (0) (0)
En,k − En,l p En − Ep
X |n(0) ; li X δHp,nk δHnl,p
|ψn,k i = (1) (1) (0) (0)
(4.7)
l6=k En,k − En,l p En − Ep

Wait: this equation has two factors of δH. Is that ok? Shouldn’t this be
a first-order correction? It is a first-order correction: dividing by the energy
difference “removes an order”.
Note that only here did we invoke the fact that δH breaks all degeneracy to
first order; otherwise dividing by the energy differences is invalid.

4.2 The general case


Joke. Sakurai’s book just says “Challenge for the experts: figure out what
happens when δH doesn’t break the degeneracy to first order.” Let’s become
experts now!

Now we don’t assume that δH breaks all degeneracy at first order: it might
still break some, but not all.
Suppose we’re again in eigenspace Vn of H (0) with dimension N . Now choose
a basis for Vn in which δH is diagonal and for which the first N 0 terms along
the diagonal of δH are all equal to Ê (1) .

10
Tony Zhang 4 February 16, 2017

So decompose Vn = V1 ⊕ V2 where V1 is the span of the first N 0 basis vectors.


To track states in V2 , use previous results. Now focusing on V1 and consider
some arbitrary member
0
N
(0)
X
(0)
|ψ i= |n(0) ; ki ak
k=1

(0)
with ak unknown. We now assume that this state and its eigenvalue evolve
power-series like as we vary λ (the standard perturbation theory assumption):

|ψiλ = |ψ (0) i + λ |ψ (1) i + . . .


En (λ) = En(0) + λEn(1) + . . .

Now assume X
|ψ (1) i = |r(0) i a(1)
r + stuff in V2
r
(1)
where the ar are unknown. Now bra-ing Equation 4.2 with hp(0) |, find that
0
N
1 X (0)
a(1)
p = (0) (0)
δHp,nk ak . (4.8)
En − Ep k=1

Now bra Equation 4.3 with hn(0) ; l| to get

0 = hn(0) ; l|(En(1) − δH)|ψ (1) i + hn(0) ; l|En(2) |ψ (0) i


(0)
= − hn(0) ; l|δH|ψ (1) i + En(2) δlk ak
(0)
X
= − hn(0) ; l| δH |pi a(1) (2)
p + En δlk ak
p
0
N
X 1 X (0) (0)
=− δHnl,p (0) (0)
δHp,nk ak + En(2) δlk ak
p En − Ep k=1
 
N 0 X 
X  δHnl,p δHp,nk (2)
 (0)
= 
(0) (0)
−En δlk 
 ak
En − Ep

k=1  p 
| {z }
Mlk

(2)
which we recognize as an eigenvalue equation. If the En are all distinct, then
(0)
we’re good! We’d have a bunch of {ak }1≤k≤N 0 that would give eigenvectors
living in V1 . This would tell us what basis we should use
Reviewing the logic:

• We take some arbitrary |ψ (0) i in V1 .


• Enforcing “trackability” and using the first order equation, we express its
first order correction in terms of itself.
• Going to second order, we [TODO TODO stuff]

11
Tony Zhang 5 February 21, 2017 (recitation)

5 February 21, 2017 (recitation)


5.1 Two-dimensional anharmonic oscillator
Consider the two-dimensional harmonic oscillator:

p2x 1 p2y 1
H0 = + mω 2 x2 + + mω 2 y 2
2m 2 2m 2
whose spectrum is obviously characterized by pairs of integers

H0 |nx , ny i = h̄ω(nx + ny + 1) |nx , ny i

since H0 is separable, whence

hxy|nx , ny i = φnx (x)φny (x).

Now perturb by
H1 = λ(x2 y + xy 2 ).
We ask what happens to the degenerate n = 1 subspace. So first we compute the
matrix elements hnk|H1 |nli. Here, n = 1 and k, l label the degenerate states.
Before continuing, we note that in the one-dimensional harmonic oscillator,
r

x= (a + a† ),
2mω
whence
r

x |0i = |1i
2mω
h̄ h̄ √
x2 |0i = (a + a† ) |1i = (|0i + 2 |2i)
2mω
r 2mω
h̄ √
x |1i = (|0i + 2 |2i)
2mω
h̄ √
x2 |1i = (3 |1i + 6 |3i).
2mω
Having invested this work, we can start bashing matrix elements:
 

h10|H1 |10i = λ h1|x2 |1i h0|y|0i + h1|x|1i h0|y 2 |0i = 0


| {z } | {z }
0 0

By symmetry in x and y, h01|H1 |01i = 0 as well.


Finally, we compute
 

h10|H1 |01i = λ h1|x2 |0i h0|y|1i + h1|x|0i h0|y 2 |1i = 0


| {z } | {z }
0 0

So the matrix elements of H1 in our degenerate block all vanish, and thus
the degeneracy is not broken to first order in perturbation theory.

12
Tony Zhang 5 February 21, 2017 (recitation)

Therefore, we go to second order. Recall that now we want to diagonalize


the matrix M (2) with entries

(2)
X (H1 )nl,p (H1 )p,nk
Ml,k = (0) (0)
p6=n En − Ep

where, in our case, n = nx + ny = 1.


Now

(2)
X (H1 )10,px py (H1 )px py ,10
M10,10 =
h̄ω(1 − px − py )
px +py 6=1

But

H1 |10i = λ(x2 y + xy 2 ) |10i


 23 
√ √ √

h̄ 
=λ (3 |1i + 6 |3i) ⊗ |1i + (|0i + 2 |2i) ⊗ (|0i + 2 |2i)
2mω
 23 
√ √ √

h̄ 
=λ 3 |11i + 6 |31i + |00i + 2 |22i + 2 |02i + 2 |20i
2mω

whence the only px , py that enter into the sum are (11, 31, 00, 22, 02, 20). Then
a little bashing tells us that

(2) 23 λ2 h̄2
M10,10 = − .
12 m3 ω 4
(2)
It turns out that M01,01 has the same value, by symmetry under coordinate
exchange of x and y.
It turns out that
λ2 h̄2
 
(2) 23 20
M =− .
12m3 ω 4 20 23

13
Tony Zhang

Part II

The hydrogen atom

14
Tony Zhang 6 February 22, 2017 (recitation)

6 February 22, 2017 (recitation)


6.1 Dirac equation
In vanilla quantum mechanics, we have the Schrödinger equation;
∂ψ
ih̄ = Hψ
∂t
2
p
where H = 2m with p = −ih̄∇, for example. Then stationary states evolve by
phase, yay. How do we make this equaion relativistic?
We know from special relativity that
p p2 1 p4
E= p2 c2 + m2 c4 ' mc2 + −
2m 8 m3 c2
where we took the p  mc limit. So why don’t we stick in p 7→ −ih̄∇ to get
p
H = −h̄2 c2 ∇2 + m2 c4

to plug into the Schrödinger equation?


There are some problems. One is the fact that we’re taking the square root
of a ∇. We can take care of that by interpreting the square root as an infinite
series.
Then we get something like

p2 1 p4
 
∂ψ
ih̄ = mc2 + − + . . . ψ.
∂t 2m 8 m3 c2

But this equation doesn’t treat time and space on the same footing, which is
bad in relativity. We want Lorentz invariance.
But what if we square the equation?
 2  
∂ ∂
ih̄ ψ = H ih̄ ψ = H 2ψ (6.1)
∂t ∂t

Now using H 2 = p2 c2 + m2 c4 , we get

∂2
−h̄2 ψ = (−h̄2 c2 ∇2 + m2 c4 )ψ
∂t2
or
∂2
   
h̄2 − c2 ∇2 + m2 c4 ψ = 0. (6.2)
∂t2
This equation is the Klein-Gordon equation. It’s nice! We recognize the
d’Alembertian operator, which is Lorentz invariant. But... there’s no way to
incorporate spin. (What if ψ has two components, one for spin up, one for down?
Then the two component equations are totally decoupled, which is unphysical.)
In fact, this equation describes spinless particles, like pions.
OK, but we like our “squared” equation 6.1. So postulate a Hamiltonian like

HD = cαi pi + mc2 β

15
Tony Zhang 6 February 22, 2017 (recitation)

where the three αi and the β are Hermitian matrices. If we require anticommu-
tation relations {αi , αj } = 2δ ij and {αi , β} = 0 and β 2 = 1, we actually satisfy
H 2 = p2 c2 + m2 c4 .
The anticommutation relations turn out to force the matrices to be 4 × 4.
One possible choice is β = diag(1, 1, −1, −1) and
 
0 ~σ
α~=
~σ 0

in block form.
Then we get the Dirac equation:
∂ψ
ih̄ = −h̄c(αi ∂i )ψ + mc2 ψ (6.3)
∂t
where ψ is now a four-component vector (actually a spinor, which is different
from a four-vector). The Dirac equation describes relativistic spin- 12 particles.
Now to look for stationary states. We have HD ψ(~x) = Eψ(~x). The solution
~
turns out to be ψ(~x) = eik·~x ξ where ξ is some spinor. Then

α · ~kh̄ + mc2 β)ξ = Eξ


(~
(h̄2 k 2 + m2 c4 )ξ = E 2 ξ
p
E = ± h̄2 k 2 + m2 c4

Since Tr α
~ = Tr β = 0, there are two positive E and two negative E.
Plotting E against p = h̄k, we find two degenerate branches. The top
corresponds to the elctron, the bottom, the positron, which just pops out.
But there’s no ground state, in the sense that if we go to high momentum,
energies get indefinitely low. How do we deal with that? We postulate that all
the negative energy levels are filled by a “Fermi sea” wait what.

16
Tony Zhang 7 February 23, 2016

7 February 23, 2016


7.1 Hydrogen review
We review the main results for the Coulomb potential in three dimensions.
These capture gross structure: zeroth-order results about the spectrum. The
Hamiltonian is:
p2 e2
H= − . (7.1)
2m r
Define the Bohr radius and the fine-structure constant:
h̄2 e2 1
a0 = 2
= 53 pm α = ≈ (7.2)
me h̄c 137
Then a basis for the set of bound states are the following orbitals:
 l  
r r Zr
ψnlm (x) = A PN e− na0 Ylm (θ, φ) (7.3)
a0 a0
where PN is some degree N = n − l − 1 polynomial.
The numbers n, l, and m are the quantum numbers.

n specifies the energy level, can be any positive integer


l specifies L2 eigenvalue, goes from 0 up to n − 1
m specifies Lz eigenvalue, goes from −l to l

Question 7.1. Shows that energy level n has degeneracy n2 .

Clearly, there’s a lot of degeneracy in this gross structure description, a


result of all the symmetry in the system.

7.2 Spin-orbit coupling


In our gross structure description, we failed to take into account the fact that
electrons are spin- 12 particles. Including spin forces us to talk about total angular
momentum
J~ = L
~ +S
~

Accordingly, we switch to a coupled basis.


Now we will specify states by n, l, j, mj , where h̄2 j(j + 1) is the eigenvalue
of J~2 and where mj is the J~z eigenvalue. A group of states with the same n, l, j
is specified in the form
nLj
where L is a letter (S, P, D, F, . . . ) corresponding to the value of l.
Using angular momentum coupling,
n=1 1S 21
n=2 2S 12 , 2P 32 , 2P 12
n=3 3S 21 , 3P 32 , 3P 12 , 3D 52 , 3D 23

17
Tony Zhang 7 February 23, 2016

Question 7.2. Why do we keep quantum number l?

Answer. You can check that L2 commutes with H, J 2 , Jz . So we still need it


in our complete set of commuting observables.

How do we couple spin to orbital angular momentum? There’s actually a


coupling term that everyone agrees with, derived from the Dirac equation.
But how would we do it? We could go into the rotating frame of the electron,
then figure out how the “orbiting nucleus” constitutes a current, which induces
a magnetic field, which interacts with the magnetic moment of the electron.
Classically, we know that magnetic moment is given as
Q ~
µ
~= L.
2M
So for an electron, insert a fudge factor g:
!
−e ~

eh̄
 ~
S eh̄
µ
~ =g S=g − = −g ~σ
2me 2me h̄ 4me
| {z }
Bohr magneton

~ field is
whence the potential energy produced by subjecting this thing to a B

~ = geh̄ ~
−µ
~ ·B ~σ · B. (7.4)
4me
Now the Dirac equation predicts that g = 2, a fact that has been experimentally
verified. But it turns out we don’t need such a large hammer to get that result.

7.2.1 Pauli equation

Consider the time-independent Schrödinger equation.

Hψ = Eψ
p2
ψ = Eψ
2m
We’d like to describe spin. So replace ψ with a two-component spinor χ and
write
p2
I2 χ = Eχ.
2m
How do we justify this? To couple spin, we might want to replace p~ 7→ ~σ · p~.
Using the fact that

(σ · ~a)(σ · ~b) = (~a · ~b)I2 + iσ · (~a × ~b), (7.5)

we get (~σ · p~)(~σ · p~) = p2 I2 , where we’re using the fact that p~ × p~ = 0.
So now we can equivalently write

(~σ · p~)(~σ · p~)


χ = Eχ. (7.6)
| 2m
{z }
HPauli

18
Tony Zhang 7 February 23, 2016

Now to couple a particle to electromagnetism, we need to replace momentum


~ where A
p~ by ~π = p~ − qc A, ~ is the electromagnetic vector potential. (We will
justify this later.)
Then we have
(~σ · ~π )(~σ · ~π )
HPauli + EM =
2m
1
= (~π · ~π I2 + i~σ · (~π × ~π ))
2m
We need to compute ~π × ~π . We have
1
(~π × ~π )k = εijk πi πj = εijk [πi , πj ].
2
But h q q i q
[πi , πj ] = −ih̄∂i − Ai , −ih̄∂j − Aj = ih̄ (∂i Aj − ∂j Ai )
c c c
whence
1 ih̄q ih̄q ih̄q ~ k = ih̄q B
~ k.
(~π × ~π )k = εijk (∂i Aj − ∂j Ai ) = εijk ∂i Aj = (∇ × A)
2 c c c c
So
ih̄q ~
~π × ~π = B. (7.7)
c
Going all the way back,
 2  
1 qA i ih̄q ~
HPauli + EM = p− I2 + ~σ · B .
2m c 2m c
The second term gives us
eh̄ ~
~σ · B,
2mc
which predicts g up to a factor of c. (The factor of c is probably the result of
differing unit conventions for the magnetic moment.)
Remark 7.3. With the Dirac equation, we’ll actually be able to go further. It
will predict nicely all the stuff that happens with spin and relativity. While the
course doesn’t strictly cover it, we’ll gain... “great insight” into it.

7.3 Dirac equation


Let’s consider adding relativity into the Schrödinger equation. From special
relativity,
E 2 = p2 c2 + m2 c4 ,
so we might expect p
H= p2 c2 + m2 c4 (7.8)
which we can Taylor expand as
r
p2
H = mc2 1+
m2 c2 !
2
1 p2 p2

2 1
= mc 1+ − + ...
2 m2 c2 8 m2 c2
p2 1 p4
= mc2 + − + ...
2m 8 m3 c2

19
Tony Zhang 7 February 23, 2016

and see that we can use the third term as a relativistic Hamiltonian perturbation.
In this case, we have p4 = p2 · p2 .
Dirac instead decided to go with the square root in Equation 7.8. So we look
for coefficients αi and β

p2 c2 + m2 c4 = (cαi pi + βmc2 )2

such that. But expanding shows that we need to have

αi2 = β 2 = 1

with α1 α2 + α2 α1 = 0. This is impossible with real numbers. But it is possible


if the αi and β are matrices. Dirac found four-dimensional matrices that worked.
Then with these αi and β,

(p2 c2 + m2 c4 )I4 = (cαi pi + βmc2 )2 .

Thus motivated, we write the Dirac equation


∂ψ
ih̄ α · p~ + βmc2 )ψ
= (c~ (7.9)
∂t
where
   
0 ~σ I2 0
α
~= β= .
~σ 0 0 −I2

Now decomposing ψ as (χ, Φ), the equation Hχ = Eχ becomes

p2 p4 1 1 dV ~ ~ h̄2
H= +V − 3 2
+ 2 2
S · L+ 2 c2
∇2 V (7.10)
2m 8m c
| {z } | {z } | 2m c r
{zdr } |8m {z }
H0 δHrel δHspin-orbit δHDarwin

In the past, we wrote λ δH for our perturbations. What’s λ here? It’ll come
out in a moment.
Let’s talk about orders of magnitude.

e2
hH0 i ∼ ∼ α2 mc2
a0
p4
hδHrel i m3 c2 p2
= p2
= = α2
hH0 i (mc)2
2m
δHspin-orbit e h̄2 1 2
= 2 2 3 e2 = α 2
hH0 i m c a0 a
0

δHDarwin
= α2
hH0 i

whence all our corrections are on the order of α2 ≈ 137−2 .

20
Tony Zhang 8 February 27, 2017 (recitation)

8 February 27, 2017 (recitation)


Recall the Dirac equation:
α · p~ + mc2 β
H = c~
where β = diag(1, 1, −1, −1) and where
 
0 ~σ
α
~= .
~σ 0

~ (We skip the


To couple in electromagnetism, we replace p~ with p~ − qc A.
derivation, mumble mumble comes from gauge invariance.) So the Hamiltonian
with electric potential φ is now
 q ~
H = c~α · p~ − A + mc2 β + qφ.
c
We’ll try to recover the fine structure of hydrogen by finding the leading
order correction in the non-relativistic limit.
H acts on four-component spinors Ψ. Write these as pairs of two-component
spinors: (χ, Φ)T . Then we find energy eigenstates:
mc2 + V
    
c~σ · p~ χ χ
= E . (8.1)
c~σ · p~ −mc2 + V Φ Φ
We can write the two components separately:
(mc2 + V − E)χ + c(~σ · p~)Φ = 0 (8.2)
2
c(~σ · p~)χ − (mc + E − V )Φ = 0 (8.3)

You may recall that the Dirac equation yields solutions for electrons and
positrons (which have negative E). But we only care about electrons for now.
So define Es = E − mc2 as the energy shift of interest.
Using the second equation, we write
1
Φ= c(~σ · p~)χ
mc2
+E−V
1
= c(~σ · p~)χ
2mc2 + Es − V
 
1 Es − V
≈ 1− c(~σ · p~)χ
2mc2 2mc2

Plugging into the first equation, we get


 
c~σ · p~ V − Es
(V − Es )χ + 1 + c(~σ · p~)χ = 0.
2mc2 2mc2
Removing the relativistic correction gives the plain old Schrödinger equation.
Now we try to commute the ~σ · p~ term on the right to the left. We have
(V − Es )~σ · p~ = (~σ · p~)(V − Es ) + [V − Es , ~σ · p~]
= (~σ · p~)(V − Es ) + σ i [V, pi ]
= (~σ · p~)(V − Es ) + ih̄∂i V σ i

21
Tony Zhang 8 February 27, 2017 (recitation)

whence commutation gives

(~σ · p~)(V − Es ) + ih̄∂i V σ i


 
c~σ · p~
0 = (V − Es )χ + c(~σ · p~)χ +
2mc2 2mc2
2 2
p~ (~σ · p~) V − Es (~σ · p~)(ih̄~σ · ∇V )
= (V − Es )χ + χ+ 2
χ+ χ
2m 2m 2mc 4m2 c2

Taking some corrections gives us the fine structure correction to the Hamil-
tonian.

22
Tony Zhang 9 February 28, 2017

9 February 28, 2017


Today, we’ll talk about hydrogen with fine-structure:
p2 p4 1 1 dV ~ ~ h̄2
H= +V − 3 2
+ 2 2
S · L + 2 2
∇2 V (9.1)
|2m{z } |8m{zc } |2m c r{zdr } |8m c{z }
H0 δHrel δHspin-orbit δHDarwin

9.1 The Darwin term


Let’s look at the first-order shifts to ` = 0 wavefunctions caused by the Darwin
term. What’s special about these states? They don’t vanish at the origin.
First, we compute
h̄2 1 e2 h̄2 3
 2
2 e
δHDarwin = ∇ − = π δ (r).
8m2 c2 r 2 m2 c2
Then
(1)
En,Darwin = hψn00 |δHDarwin |ψn00 i
π e2 h̄2
= |ψn00 (0)|2
2 m2 c2
π e2 h̄2 1
=
2 m2 c2 πa30 n3
1 1
= α4 mc2 3
2 n

Question 9.1. Why is it OK to use nondegenerate perturbation theory here?

Answer. We’re only considering ` = 0 states (mumble).

How do we interpret the Darwin term? It accounts for the fact that the

electron behaves like a little sphere with the size of its Compton wavelength me .
Let’s see how that works.
The potential felt by the particle isn’t V (r) anymore. We need to average
over the entire sphere:
Z
Ṽ (r) = V (r + ρ)f (ρ) dρ (9.2)
R
where f (ρ) gives a normalized “mass density”: f (ρ) dρ = 1. If f (ρ) = δ(ρ),
we recover Ṽ = V .
So we expand with a Taylor series:
X 1X
V (r + ρ) = V (r) + (∂i V )(r)ρi + (∂i ∂j V )(r)ρi ρj + . . . (9.3)
i
2 i,j

which we can substitute:


Z X Z
3
Ṽ (r) = V (r)f (ρ) d ρ + (∂i V )(r) ρi f (ρ) d3 ρ
i
Z
1X
+ (∂i ∂j V )(r) ρi ρj f (ρ) d3 ρ + . . .
2 i,j

23
Tony Zhang 9 February 28, 2017

Clearly, the zeroth-order term factors nicely and normalization kicks in. Now
assume f is spherically symmetric. Then by symmetry, the first-order term
vanishes. Moreover, the second-order term does as well unless i = j:
Z
1X 2
Ṽ (r) = V (r) + (∂i V )(r) ρ2i f (ρ) d3 ρ + . . .
2 i
Z
1
= V (r) + (∇2 V )(r) ρ2 f (ρ) d3 ρ.
6

We will now assume


(
3 1 ρ < ρ0
f (ρ) = . (9.4)
4πρ30 0 ρ > ρ0

Then we can evaluate


ρ0
4πρ2 dρ ρ2
Z Z
3
ρ2 f (ρ)d3 ρ = 4 3
= ρ20 (9.5)
0 3 πρ0
5

whence, setting the radius ρ0 to the Compton wavelength mc ,

h̄2
 
1
Ṽ (r) = V (r) + ∇2 V . (9.6)
10 m2 c2
1
This correction is very close to the actual Darwin term (which has prefactor 8
1
instead of 10 ).

9.2 The relativistic correction


We have
(1) 1
En,rel = − hψnlm |p4 |ψnlm i (9.7)
8m3 c2
where p4 = (p · p)2 . Because [p4 , L2 ] = [p4 , Lz ] = 0, our lemma from before tells
us that the |nlmi are good states to do perturbation theory with.
Using Hermiticity of p2 (discussed on pset 3),

(1) 1
En,rel = − hp2 ψnlm |p2 ψnlm i .
8m3 c2

Now the Schrödinger equation tells us that

p2 ψ = (E (0) − V )2mψ,

whence
(1) 1
En,rel = − h(En(0) − V )ψ|(En(0) − V )ψi
2mc2
1
=− hψ|(En2 − 2En V + V 2 )|ψi
2mc2
1
=− [(En(0) )2 − 2En(0) hV inlm + hV 2 inlm ]
2mc2

24
Tony Zhang 9 February 28, 2017

Question 9.2. Why couldn’t we have applied p2 twice on one side instead of
invoking Hermiticity and putting one p2 on each side?

Answer. p2 doesn’t commute with V !

Now we recall the virial theorem:

Theorem 9.3 (Virial theorem)


hV i = 2En (9.8)

Remark 9.4 (Virial theoerm mnemonic). Imagine the En < 0. The kinetic
energy must counteract this binding energy, so K = hT i = −En > 0. Then
hV i = En − hT i = 2En .

Then we can continue


(1) 1
En,rel = − [(En(0) )2 − 4(En(0) )2 + e4 hr−2 inlm ]
2mc2

We computed
1 4n
hr−2 i = = (En(0) )2 (9.9)
a20 n3 (` + 21 ) l + 12

on a previous pset.

Question 9.5. Verify the second equality in Equation 9.9.

So we have
(0)
(En )2
   
(1) 4n 1 4 2 1 4n
En,rel = − − 3 = − α mc − 3 (9.10)
2mc2 ` + 21 8 n4 ` + 21

Now this result was derived in the uncoupled basis |n`m` ms i. But it only
depends on n and `. Therefore, linear combinations of state with the same n
and ` experience the same relativistic energy shift.
So the result is also valid in the coupled basis |n`jmj i.

9.3 Spin-orbit coupling


We have
e2
 
1
(1) ~ ~

En,so = n`jm 3 S · L n`jm .

2m2 c2 r

~ ·L
Using 2S ~ = J 2 − S 2 − L2 , and the known result of hr−3 i, we get

(0)
(1) (En )2 n(j(j + 1) − `(` + 1) − 43 )
En,so =
m2 c2 `(` + 12 )(` + 1)

for ` 6= 0 and 0 for ` = 0. Interestingly, letting ` → 0 here doesn’t give 0.

25
Tony Zhang 9 February 28, 2017
1
Joke. If you want to find the expectation value of r3 and you have to go to the
Laguerre polynomials, you’re finished.

Finally, adding the two energy shifts, we get the total shift of the terms
we’ve considered today:

(En )2 j(j + 1) − 3`(` + 1) − 43


 
(1) (1)
Erel + Eso = 3 + 2n (9.11)
2mc2 `(` + 21 )(` + 1)

This result gives the fine-structure correction for ` 6= 0 states, since they have no
Darwin shift. But it’s valid even for ` = 0 states as well, despite the spin-orbit
term being wrong! Indeed, the incorrect spin-orbit term in the ` → 0 limit
exactly compensates for the shift from the Darwin term that we didn’t explicitly
add in.
Yay, so we have a nice formula that depends on j and `. Are we done? Not
quite. Because `⊗ 12 = (`+ 12 )⊕(`− 12 ), we always have ` = j ± 12 . Plugging these
values into Equation 9.11, we get the same expression both times. Therefore, we
can write the grand formula for the fine structure energy shift of hydrogen:

(0)
(En )2
   
(1) 4n 4 2 1 n 3
Efs =− − 3 = −α mc − . (9.12)
2mc2 j + 21 2n4 j + 1
2
4

9.4 Remarks on Zeeman


Now we have the nice fine-structure picture. What happens when we stick in a
weak magnetic field? Then we’ll get energy level splittings within the remaining
degenerate subspaces, and the gaps will be exactly equal. It’s a very nontrivial
result.

26
Tony Zhang 10 March 1, 2017 (recitation)

10 March 1, 2017 (recitation)


Remember we were on this “epic quest” to get the relativistic correction to the
Schrödinger equation from the Dirac equation.

10.1 Fine structure from Dirac


We started wth the Dirac equation:

HD Ψ = EΨ,

with four-component spinor Ψ and Dirac Hamiltonian

α · p~ + βmc2 + V.
HD = c~

We decomposed Ψ = (χ, Φ) and found that


 
1 V − Es
Φ' 1+ c~σ · p~χ.
2mc2 2mc2

Last time, we got to


 2
p4

p ih̄
+V − + (~
σ · p
~ )(~
σ · ∇V ) χ = Es χ.
2m 8m3 c2 4m2 c2

(Imagine χ as (χ↑ , χ↓ ).)


Now let’s compute

(~σ · p~)(~σ · ∇V ) = σ i pi σ j ∂j V
= (δ ij + iεijk σ k )pi ∂j V
p · ∇V ) + iεijk pi (∂j V )σ k
= (~

We commute pi through ∂j V using

pi ∂j V = ∂j V pi + [pi , ∂j V ] = ∂j V pi − ih̄∂i ∂j V.

The latter term is symmetric in i and j, so contractingit with the Levi-Civita


tensor gives 0. Therefore, we have

p · ∇V ) − iεijk (∂i V )pj σ k .


(~σ · p~)(~σ · ∇V ) = (~

Now assume a central potential, since we’re trying to get at the relativistic
correction for hydrogen. Then
∂r ri
∂i V = V 0 (r) = V 0 (r)
∂ri r
so that
V 0 (r) j k
(~σ · p~)(~σ · ∇V ) = p~ · ∇V − iεijk ri p σ
r
V 0 (r) ~
= p~ · ∇V − i L · ~σ
r

27
Tony Zhang 10 March 1, 2017 (recitation)

Plug this expression into our equation to get


 
 p2 p4 ih̄ h̄ V 0 (r) ~ ~σ 
 + V − + (~
p · ∇V ) + L·  χ = Es χ.
 2m 8m3 c2 4m2 c2 2m
|
2 c2 r
{z 2}
Hso

where we recognize the spin-orbit coupling term:

1 V 0 (r) ~ ~ e2 ~ · S.
~
Hso = L · S = L
2m2 c2 r 2m2 c2 r4

The term containing p~ · ∇V is... not Hermitian. That’s rather unfortunate,


but it isn’t a mistake. To see, we’ll use the interpretation of χ as half of spinor Ψ.
In the Dirac equation, the probability measure of being somewhere is

Ψ† (x)Ψ(x) d3 x.

The integral of this quantity is unity, which is obviously conserved. The integral
of χ† χ, however, isn’t necessarily conserved.
But we can also write
Z Z
Ψ† Ψd3 r = (χ† χ + Φ† Φ)d3 r
Z  
3 † 1 †
' d r χ χ+ ((~σ · p~)χ) (~σ · p~)χ
4m2 c2
2
Z  
3 † † p
= d r χ χ+χ χ
4m2 c2
p2
Z  
= χ† 1 + χ d3 r
4m2 c2

which is conserved in time.


From here, we can extract a two-component spinor whose integral is con-
served:
p2
 
χs = 1 + χ,
8m2 c2

Writing χ in terms of χs gives gives

h̄2
HD = ∇2 V.
8m2 c2

TODO TODO TODO TODO TODO

28
Tony Zhang

Part III

The WKB approximation

29
Tony Zhang 11 March 2, 2017

11 March 2, 2017
11.1 WKB; the semiclassical approximation
Today, we look at the Wentzel-Kramers-Brillouin, or WKB, approximation.
This approximation looks at the limit when quantum effects get small. There
are two ways to quantify that:

• The de Broglie wavelength λ becoming small (λ  relevant length scale)


Recall that λ = hp .

• h̄ → 0. In this case, we’ll need another quantity with dimensions of action


to compare h̄ with.

First, let’s decompose a complex wavefunction as a magnitude and phase:


p iS(x,t)
Ψ(x, t) = ρ(x, t)e h̄ .

Here S(x, t) is real with dimensions of action. ρ(x, t) is just the probability
density |Ψ(x, t)|2 .
How do we interpret this phase? We recall the probability current:

J~ = =(Ψ∗ ∇Ψ).
m
Let’s write it in terms of ρ and S.
Take the gradient of Ψ:
1 ∇ρ iS √ i iS
∇Ψ = √ e h̄ + ρ ∇Se h̄
2 ρ h̄

whence
1 i
Ψ∗ ∇Ψ = ∇ρ + ρ∇S
2 h̄
and
∇S
J~ = ρ . (11.1)
m
Now recall from classical electrodynamics that

p~
J~ = ρ~v = ρ ,
m
so we think of ∇S ∼= p~ as some sort of momentum. Admittedly, this equivalence
is vague, but for a free particle, it makes sense:
~·~
p x−Et
Ψ(x, t) = ei( h̄ )

Then we see that S = p~ · x − Et and ∇S = p~.

30
Tony Zhang 11 March 2, 2017

11.2 Local momentum, de Broglie wavelength


We now go to one-dimension. Consider a particle in some energy eigenstate
under some potential. It evolves in time by phase:
iEt
Ψ(x, t) = Ψ(x)e− h̄ .

It must satisfy the time-independent Schrödinger equation:

h̄2 00
− ψ + V (x)ψ = Eψ.
2m
Reorganize to define a local notion of “momentum”:

−h̄2 ψ 00 = 2m (E − V (x)) ψ = p2 (x)ψ


| {z }
p2 (x)
kinetic energy 2m

Importantly, p(x) is a vanilla function, not a quantum operator. We can also


define a local de Broglie wavelength by
h
λ(x) = ,
p(x)
where we note we’re using regular h, not h̄.
Note that in a “classically forbidden region” (E < V (x)), p(x) = ±iκ(x) for
some real κ.

11.3 Solving the equation


Now we posit
iS(x)
ψ(x) = e h̄

with S now a complex quantity with dimensions of action. Allowing S to be


complex means there is no loss of generality. Solving the time-independent
Schrödinger equation,

iS d2 iS
p2 (x)e h̄ = −h̄2 (e h̄ )
dx2
d iS 0 iS

= −h̄2 e h̄
dx h̄
 00
(S 0 )2 iS

iS iS
= −h̄2 e h̄ − 2 e h̄
h̄ h̄
whence we get the following:

p2 (x) = (S 0 )2 − ih̄S 00 (11.2)

This equation is exact: we haven’t yet made any approximations. It’s


also pretty nice, though it doesn’t look that way at first: it’s nonlinear and
second-order.
However, consider what happens if potential V is constant: the function
p2 (x) = 2m(E − V ) is constant as well. But then S 0 = p, so S 00 = 0, so the
second-order term dies.

31
Tony Zhang 11 March 2, 2017

In general, for a sufficiently slowly varying potential, ih̄S 00 is small. What do


we mean by “sufficiently slowly”? It’s slow if, over the de Broglie wavelength of
the particle we’re considering, the potential is practically constant. The “small”
quantity here is thus wavelength λ = 2πh̄ p . So we can imagine expanding S (the
proxy for our wavefunction) in terms of λ.
But λ isn’t some fixed parameter! Instead, we can expand in h̄, which shows
up in λ:
S = S0 (x) + h̄S1 (x) + O(h̄2 ). (11.3)
The first order truncation is the WKB approximation.
Plugging into Equation 11.2,

(S00 + h̄S10 + . . . )2 − ih̄(S000 + h̄S100 + . . . ) = p2 (x).

Keeping only terms to first order per the WKB approximation,

((S00 )2 − p2 (x)) + h̄(2S00 S10 − iS000 ) + O(h̄2 ) = 0. (11.4)

Now we want this “power series” in h̄ to be identically zero. Therefore, the


zeroth order equation gives S00 = ±p(x), and, for some arbitrary x0 ,
Z x
S0 = ± p(x0 ) dx0 . (11.5)
x0

The first order term gives 2S00 S10 = iS000 . We can rearrange:

i S000 i ±p0 i p0
S10 = = =
2 S00 2 ±p 2p

and integrate to solve for S1 :


i
S1 = ln p(x) + C.
2

Therefore, to first order in h̄, we can find our wavefunction as


p(x0 ) dx0 − 12 ln p(x) C
Rx
iS0 (x)
+
ih̄S1 (x)
± h̄i
ψ(x) ≈ e h̄ h̄ =e x0
e e .

A general solution is just a superposition of the two solutions we get from taking
either plus or minus. Depending on whether E > V (x) (classically allowed) or
otherwise (classically forbidden),

i x i x
 Z   Z 
A B
ψ(x) = p exp + p(x0 ) dx0 + p exp − p(x0 ) dx0
p(x) h̄ x0 p(x) h̄ x0
(11.6)
 Z x   Z x 
A 1 0 0 B 1 0 0
ψ(x) = p exp + κ(x ) dx + p exp − κ(x ) dx
κ(x) h̄ x0 κ(x) h̄ x0
(11.7)

The coefficients A and B parameterize our solution family. How valid are these
approximations?

32
Tony Zhang 11 March 2, 2017

11.4 Validity
Because we have a power series expansion in h̄ in Equation 11.4, we expect
it the first-order term to be smaller than the zeroth-order one. Therefore, we
expect:

|h̄S00 S10 |  |S00 |2


|h̄S10 |  |S00 |
0
h̄p
p  |p|

2πh̄
But we recognize the de Broglie wavelength λ = p , whence, ignoring constant
factors,
λ(x) dp  |p|.

(11.8)
dx

Alternatively, we can write


0
h̄p
p2  1

 
d h̄
dx p  1

whence

λ  λ (11.9)
dx

These two results show that local momentum and de Broglie wavelength
change are basically constant over the length scale of a de Broglie wavelength.
That is, the amount they change is much less than their values.

11.5 The connection formula


The WKB wavefunction approximations we just derived are really only valid
when V (x) > E or otherwise. The approximation breaks down at turning points,
where we go from classically allowed to classically forbidden. Indeed, p(x) and
κ(x) go to zero, and wavefunction ψ diverges.
dp
The reason is that at turning points, λ(x) dx becomes comparable to or
greater than p, which violates the semiclassical approximation.
Our problem is now as follows: we have good approximations of the energy
eigenstate in the classically allowed region and in the classically forbidden region.
However, these two approximations become invalid near turning points. How
do we “stitch” them together– “connect” them– to get one wavefunction?
Let’s say the potential becomes classically disallowed at a. For x  a,
E > V , and we can use WKB. Similarly for x  a. However, how do we connect
the two solutions at the turning point? We can’t do the 8.04 thing of enforcing
continuity because the two WKB solutions, in general, disagree at x = a.

33
Tony Zhang 11 March 2, 2017

The connection formula says that the solutions in the two regimes take
the form
 Z a   Z a 
2A 1 π B 1 π
p cos p(x0 ) dx0 − −p sin p(x0 ) dx0 − xa
p(x) h̄ x 4 p(x) h̄ x 4
(11.10)
 Z x   Z x 
A 1 B 1
p exp − κ(x0 ) dx0 +p exp κ(x0 ) dx0 xa
κ(x) h̄ a κ(x) h̄ a
(11.11)

These connect at a.
Look at an example first. Suppose our potential is infinite for x ≤ 0, and
grows linearly-ish for x > 0. Moreover, suppose V (a) = E, with x > a classically
forbidden.
The WKB approximation in the classically allowed region, enforcing boundary
conditions at the origin, is:
 Z x 
1 1
ψ(x) = p sin p(x0 ) dx0
p(x) h̄ 0
| {z }
φ(x)

where we define φ(x), require φ(0) = 0, and ignore normalization.


But we need the solution to take the form of 11.10. Write φ(x) as follows:

1 a
Z  Z a 
π 1 π
φ(x) = p(x0 ) dx0 − − p(x0 ) dx0 −
h̄ 4 h̄ x 4
| 0 {z } | {z }
∆ ξ(x)

Joke. We’re feeling generous today, so we add and subtract whatever we want!

Then
1
ψ(x) = p sin φ
p(x)
1
=p sin(∆ − ξ(x))
p(x)
1 1
=p sin ∆ cos ξ − p cos ∆ sin ξ
p(x) p(x)

TODO TODO TOD TODO TODO

34
Tony Zhang 12 March 6, 2017 (recitation)

12 March 6, 2017 (recitation)


12.1 Gamow’s theory of α-decay
α radiation is very mysterious, since nuclei lifetimes vary by 25 orders of
magnitudes. We’ll see how we can understand it as quantum tunneling.
Consider a nucleus of charge Z + 2, with an α-particle (a helium nucleus).
By conservation of energy, we have

mZ+2 c2 = mZ c2 + mα c2 + E

with E the energy of the escaping α particle.


Gamow imagined the α particle as a particle in a potential well of finite height
(within some radius r1 in which the strong force keeps the nucleus together).
Outside r1 , the potential is just a Coulomb potential:
(
−|V0 | r < r1
V (r) = 2Ze2 (12.1)
r r > r1

We’ll use the WKB approximation here. Generally r1 is really tiny, so the
potential barrier is tall and wide. Then consider an α particle inside the barrior,
with energy E > 0 to allow tunneling. Let r2 denote the turning point, so that

2Ze2
= E.
r2

Since the barrier is tall and wide, we can derive a transmission coefficient
T ∼ e−γ with
s
1 r2 1 4me2 Z r2 r2
Z Z r
0 0
γ= κ(x ) dx = − 1 dr
h̄ r1 h̄ r2 r1 r
√  r  r2
2mE p r
= r(r2 − r) + r2 sin−1
h̄ r2 r=r1

But r1  r2 , so we can simplify


√  
2mE 1 √
γ= πr2 − r1 r2
h̄ 2
Z p
= k1 √ − k2 Zr1
E
in experimentalist-friendly form, where

π 2me2 1 1
k1 = = 1.98 (MeV) 2 k2 = 1.49 (fm)− 2 .

Now the tunneling probability is is therefore |T |2 = e−2γ . Imagining the


alpha particle rattling inside the nucleus at velocity v, the lifetime od the nucleus
should be
v −2γ
Γ = Γ0 e−2γ = e .
2r1
1
Plotting experimental results for log Γ
Γ0 against E − 2 gives a line, as expected!

35
Tony Zhang 13 March 7, 2017

13 March 7, 2017
First: notation change. We use k(x) instead of p(x), where p = h̄k. Therefore,
we write
2m
k 2 (x) = (E − V (x))
h̄2
2m
κ2 (x) = 2 (V (x) − E)

Then the condition |h̄p0 |  |p2 | becomes |k 0 |  |k 2 |, and the WKB exponents
become nicer. The probability density and current become nicer too.
For a WKB wavefunction of the form
 Z x 
A 0 0
Ψ(x) = p exp i k(x ) dx ,
k(x) a

we would have
|A|2
ρ=
k(x)
and probability current
h̄ h̄
J= =(Ψ∗ Ψ0 ) ' |A|2
m m
We write down the connection formulae. If a is a turning point from allowed
to forbidden,
 Ra Ra
 √2A cos x k(x0 ) dx0 − π4 − √B sin x k(x0 ) dx0 − π4
 
x<a
k(x) k(x)
ψ(x) = Rx x
√ A
exp − a κ(x0 ) dx0 + √ B
exp a κ(x0 ) dx0
 R 
x>a
κ(x) κ(x)
(13.1)
Similarly, from forbidden to allowed at b:
  R  R 
 √ A exp − xb κ(x0 ) dx0 + √ B exp xb κ(x0 ) dx0 x<b
κ(x) κ(x)
ψ(x) =
 √2A cos x k(x0 ) dx0 − π − √B sin x k(x0 ) dx0 − π
R  R 
b 4 b 4 x>b
k(x) k(x)
(13.2)

13.1 Establishing the connection formulae


Let’s suppose that we’re going from classially allowed to classially forbidden at
point a. Since we assume V varies slowly, suppose it is well approximately by a
linear potential near a:
V (x) − E = g(x − a) (13.3)
for g > 0.
Far from the turning point, our wavefunction lives in the two-dimensional
function space specified by the WKB approximation:
 Z x  Z x 
A 0 B 0
ψ(x) = p exp − κ dx + p exp κ dx (13.4)
κ(x) a κ(x) a
 Z a   Z a 
C D
ψ(x) = p exp i k dx0 + p exp −i k dx0 (13.5)
k(x) x k(x) x

36
Tony Zhang 13 March 7, 2017

Using our linearized potential, we have


2mg
k2 = (a − x) x≤a (13.6)
h̄2
2mg
κ2 = 2 (x − a) x≥a (13.7)

With our linearized potential, the Schrödinger equation

h̄2 00
− ψ + (V (x) − E)ψ = 0
2m
becomes
h̄2 00
− ψ + g(x − a)ψ = 0.
2m
But we don’t like dimensionful quantities, so let’s get rid of them. We can get a
characteristic length scale
 2  13

L=
mg
h̄2
using mL2 = gL, both sides of which have dimensions of energy.
Now scale x as   31
2mg
u= (x − a) = η(x − a)
h̄2
where we define η. Then we get Airy’s differential equation

d2 ψ
− uψ = 0, (13.8)
du2
whose solutions are linear combinations of the Airy functions, cousins to the
Bessel functions. Asymptotically, the Airy functions look like
1
(
√1
|u|− 4 e−ζ u1
Ai(u) ' 21 π − 1 (13.9)
√ |u| 4 cos ζ − π

π 4 u0
1
(
√1 |u|− 4 eζ u0
π
Bi(u) ' 1 − 1 π
 (13.10)
− √π |u| 4 sin ζ − 4 u  −1
3
where ζ = 32 |u| 2 .
Remark 13.1. Whoever determined these asymptotic forms of the Airy func-
tions have really done a lot of the work for us. Near the turning point, the
solutions are Airy functions Ai and Bi, and it just so happens that they take
WKB form asymptotically!
1 1
To see: We have u ∼ (V − E) ∼ k 2 , whence |u|− 4 ∼ k − 2 . Similarly, the
3
exponentials have exponents ζ, √
which go like u 2 . But the exponents in the
WKB solutions are integrals of x (using our linearized potential), giving us
our 32 powers.
Remark 13.2. Consider the asymptotic form of Ai:
1 2
3
ψ(u) = 1 e− 3 u 2 .
u 4

37
Tony Zhang 13 March 7, 2017

It actually exactly solves the equation


d2 ψ
 
5 1
− u + ψ = 0.
du2 16 u2
So far from the turning point, u2 is big and this equation just becomes Airy’s
differential equation (Equation 13.8).
The moral is that the asymptotic forms of the Airy functions approximate
both the solution near the turning point (by approximately satisfying Equa-
tion 13.8) and the WKB solutions far from the turning point (because it takes
the WKB form).

Now we write:
k 2 = η 3 (a − x) = −η 2 u = η 2 |u| u<0 (13.11)
2 3 2
κ = η (x − a) = η u u>0 (13.12)
We’re now in a position to ask: how good are the Airy asymptotic forms as
approximations to the wavefunctions? Well, we want |k 0 |  |k 2 |.
3 1
Question 13.3. Show that this condition is equivalent to |u| 2  2
, whence
|u|  0.63.

Remark 13.4. We’re actually trying to satisfy a double constraint: we want a


region where |u| is large enough for the WKB approximation to hold, and where
|u| is small enough so the potential can be linearized. This overlap is important
because we will be treating the WKB approximations away from the turning
point as asymptotic forms of Airy functions.

Our program now is to choose coefficients for Equations 13.4 and 13.5 that
make them match up asymptotically with some linear combination of Ai and Bi.
Evaluate integrals near the turning point using the linearized potential
Z x Z x
0 0 3 1 2 3
κ(x ) dx = η 2 (x0 − a) 2 dx0 = |u| 2 = ζ (13.13)
3
Za a a

k(x0 ) dx0 = ζ (13.14)


x

and WKB prefactors


1 1 1
p =p = 1 1
κ(x) k(x) η 2 |u| 4

Now we rewrite the solutions away from the turning point (Equations 13.4
and 13.5) as
1
A|u|− 4 −ζ B 1
u0 ψ(x) = √ e + √ |u|− 4 eζ
π π
' 2A Ai(u) + B Bi(u) (13.15)
− 14 −iζ
|u| e 1
u0 ψ(x) = C √ eiζ + D √ |u|− 4
π π
 iπ iπ
  iπ iπ

' Ce 4 + De 4 Ai(u) + iDe− 4 − iCe 4 Bi(u)

(13.16)

whence matching coefficients gives the connection formula.

38
Tony Zhang 13 March 7, 2017

Question 13.5. Derive Equation 13.16.

13.2 Arrows and asymmetric validity


We note that in the connection formulae, we rarely have both A and B present,
in the sense that we usually have an idea of what the wavefunction looks like
in one region. For example, in a forbidden region unbounded to the right, we
know that the exponential can only be decaying.
Let’s flesh out that remark a little bit more. Suppose we know a priori that
the wavefunction in some forbidden region is exponentially decaying. Then we
see from the connection formula (Equation 13.1) that B = 0 exactly— there can
be no exponentially growing term.
As a result, we know that there is a term of the form
Z a 
2A π
p cos k(x0 ) dx0 −
k(x) x 4

in the classically allowed region.


Arrows:
Z a   Z x 
2A π A
p cos k(x0 ) dx0 − ⇐p exp − κ(x0 ) dx0
k(x) x 4 κ(x) a
Z a  Z x 
B π B
− p sin k(x0 ) dx0 − ⇒p exp κ(x ) dx 0 0
k(x) x 4 κ(x) a

13.3 Tunneling
Because *mumble mumble arrows*, in a potential barrier in a tunneling situation,
we can assume that with scattering to the right, with incident, reflected, and
transmitted amplitudes A, B, F , the wavefunction is almost completely decaying.
Then the tunneling probability is just the ratio of probability fluxes

|F |2
T =
|A|2

(where we’ve cancelled factors of m ).
But WKB tells us that this quantity is just the amount of decay introduced
by the exponential decay:
T = e−2Θ
where Z b
Θ= κ(x0 ) dx0
a

39
Tony Zhang 14 March 8, 2017 (recitation)

14 March 8, 2017 (recitation)


14.1 Tunneling
Let’s talk more about tunneling. Consider a potential barrier off which we
scatter incoming Aeikx from the left to get reflected Be−ikx and transmitted
F eikx . There is no e−ikx component to the right of the barrier.
Assuming a tall, broad barrier, we expect |F |  |A|, and that the wavefunc-
tion in the forbidden region will be mostly decaying.
Fix energy E, and take turning points as a < b. We’ll connect at b from the
transmitted region. For x  b,
 Z x 
F i iπ
ψ(x) = p exp p(x0 ) dx0 −
p(x) h̄ b 4
 Z x  Z x 
F p 0 π p 0 π
=√ cos dx − − i sin dx −
p b h̄ 4 b h̄ 4
where we add the phase for convenience.
Now posit solution
1 x 1 x
  Z   Z 
1 0 0 0 0
ψ(x) = p C exp − κ(x ) dx + D exp + κ(x ) dx
κ(x) h̄ a h̄ a
"Z # " Z # !
b b
1 κ −θ κ θ
=√ C exp e + D exp − e
κ x h̄ x h̄

for a  x  b, where
1 b
Z
θ= κ dx0
h̄ a
Remember our expectation that in the forbidden region, the wavefunction is
mostly decaying? We’ll check that C  D to verify.
Indeed, connection gives
1 −θ
D= Fe C = −iF eθ
2
whence D  C by a factor of e2θ , which incidentally will scale as tunneling
lifetime. We really don’t have a right to keep D, (author’s sketchy note: since
the correction it provides is smaller than the O(h̄) corrections we drop in making
the WKB approximation?) so basically we assume that the wavefunction is
purely exponentially decaying.
Another connection into the region left of the barrier will give |A| = |B| = |C|,
whence
|F | |F |
= = e−θ
|A| |C|
and
|B|
= 1.
|A|
The transmission probability is therefore
|F |2 |B|2
T = = e−2θ R= =1
|A|2 |A|2
They don’t sum to 1 because of approximation errors.

40
Tony Zhang

Part IV

Time-dependent perturbation
theory

41
Tony Zhang 15 March 9, 2017

15 March 9, 2017
15.1 Time-dependent perturbation theory
Today we take up time-dependent perturbation theory. The setup is
simple. Given a known, time-independent H (0) , we want to understand

H(t) = H (0) + δH(t).

Now we shouldn’t think of a spectrum, since there really is no longer a concept


of energy eigenstates. Instead, we just want to find how states evolve |ψ(t)i.
We want to think of δH as having finite time support. So we start at H (0) ,
turn on a perturbation, and then turn it off. The perturbation should not last
long.

15.2 The interaction picture


Today we’ll work in the interaction picture, a hybrid of the Schrödinger and
Heisenberg pictures. Recall that given Schrödinger observable AS , we can define
a corresponding Heisenberg observable

AH = U † AS U

where U is the unitary time evolution operator. Taking the t expectation of AS


is the same as taking the t = 0 expectation of AH .
Remark 15.1. You should think of U † as an operator that “brings a state to
rest”:
U † |Ψ(t)i = U † U |Ψ(0)i = |Ψ(0)i .

OK, now consider the unitary time-evolution operator for H (0) :


iH (0) t
UH0 = e− h̄

and define
iH (0) t
|Ψ̃(t)i = U † |Ψ(t)i = e+ h̄ |Ψ(t)i . (15.1)
(0)
We’re “bringing |Ψ(t)i to rest” with respect to H .
We’d like a time-dependent Schrödinger equation for |Ψ̃(t)i. Start with the
familiar one:
ih̄∂t |Ψ(t)i = (H (0) + δH(t)) |Ψ(t)i
Then

ih̄∂t |Ψ̃(t)i = ih̄∂t (UH 0
|Ψ(t)i)
= ih̄(∂t U † ) |Ψ(t)i + ih̄U † ∂t |Ψ(t)i
iH (0) t
= −H (0) |Ψ(t)i + e h̄ (H (0) + δH(t)) |Ψ(t)i
= −H (0) |Ψ(t)i + U † (H (0) + δH(t))U |Ψ̃(t)i
= (U † δH(t) U ) |Ψ̃(t)i .

Thus, we’re motivated to define



δH(t)
f = UH 0
δH(t)UH0 (15.2)

42
Tony Zhang 15 March 9, 2017

so that

ih̄ |Ψ̃(t)i = δH(t)
f |Ψ(t)i . (15.3)
∂t
So now we have

• “time dependent states” |Ψ̃(t)i


f with respect to H (0) .
• Heisenberg operator δH

Once we solve this problem, we can add in the H (0) time dependence simply as
iH (0) t
|Ψ(t)i = e− h̄ |Ψ̃(t)i .

15.3 Exact evolution in the interaction picture


Now let’s introduce a basis: the eigenbasis |ni of H (0) . It’s a perfectly good
orthonormal basis, so let’s expand |Ψ̃(t)i:
X
|Ψ̃(t)i = cn (t) |ni . (15.4)
n

for unknown coefficient functions cn (t). Then we can bop this equation with
UH (0) to get X iEn t
|Ψ(t)i = cn (t)e− h̄ |ni . (15.5)
n

This equation should make you happy! If the cn (t) were constant, we recover
ordinary time-evolution of |Ψ(t)i under H (0) .
Now plugging into our Schrödinger equation,
X X X XX
ih̄ċm (t) |mi = cn |mi hm| δH
f |ni = |mi hm|δH|ni
f cn (t)
m n m m n

where we have matrix elements


hm|δH|ni
f f mn = e i(Emh̄−En )t hm|δH|ni
= δH (15.6)
| {z }
eiωmn t

Then we get a series of coupled differential equations for the cm :


X X
ih̄ċm (t) = (δH)
f mn cn (t) = eiωmn t (δH)mn (t)cn (t) (15.7)
n n

15.3.1 Example: two-state system

Problem 15.2. Let’s start with a two-state system with stationary states
|ai , |bi. with respective energies Ea , Eb . Define a charactristic frequency
Ea − Eb
ω0 = .

Remember how we said perturbations shouldn’t last long? We’ll take the
shortest possible perturbation: a delta function! We have δH = U δ(t) for
Hermitian U with Uaa = Ubb = 0 and Uab = α.
Suppose the system is in |ai for t = −∞. What is the probability we find it
in |bi at t = ∞?

43
Tony Zhang 15 March 9, 2017

Start by considering

|Ψ̃i = ca (t) |ai + cb (t) |bi .

The initial conditions are

ca (t) = 1 cb (t) = 0

for t < 0. The coefficients satisfy

ih̄ċa = eiωaa t (δH)aa ca (t) + eiωab t (δH)ab cb (t)


ih̄ċb = eiωba t (δH)ba ca (t) + eiωbb t (δH)bb cb (t)

or

ih̄ċa = eiω0 t αcb (t)δ(t)


ih̄ċb = e−iω0 t α∗ ca (t)δ(t)

Now recall that we generally write

f (x)δ(x) = f (0)δ(x).

Can we do that here?


Let’s split that into two steps:

ih̄ċa ' αcb (t)δ(t) ' αcb (0)δ(t).

The first step is relatively kosher, since eiω0 t is continuous. But in general we
have no idea whether the ca and cb are continuous. Indeed, we really don’t
expect them to be. So it’s sketchy to make the second step.
Therefore, we write

ih̄ċa = αcb (t)δ(t)


ih̄ċb = α∗ ca (t)δ(t)

To model the delta function, make the substitution

δ(t) 7→ ft0 (t)


1
where ft0 is t0 for 0 < t < t0 and 0 otherwise. Sending t0 → 0 recovers the delta
function.
Now for t ∈ [0, t0 ],
i 1
ċa = − α cb (t)
h̄ t0
iα∗ 1
ċb = − ca (t)
h̄ t0

Then
iα |α|2
c̈a = − ċb = − 2 2 ca (t)
h̄t0 h̄ t0
and something similar for cb , whence
|α|t |α|t
ca = β0 cos + β1 sin
h̄t0 h̄t0

44
Tony Zhang 15 March 9, 2017

so that initial conditions require β0 = 1 and β1 = 0. We can use ċb ∼ ca to find


cb . In particular, we will find that

|α|t
ca (t) = cos
h̄t0
i|α| |α|t
cb (t) = − sin
α h̄t0
So at t0 , we have

|α|
ca (t) = cos

i|α| |α|
cb (t) = − sin
α h̄
All t0 dependence has now dropped out, so taking the limit of regularizer t0 → 0
is trivial, and we get
|α|
Pr(b; t = 0+ ) = sin2

Remark 15.3. One might argue that introducing the step function regulation
is unphysical. On the contrary, the regulation is totally the right thing to do;
the problem itself was unphysical!

15.4 Perturbative solution


Suppose we write
ih̄∂t |Ψ̃(t)i = λ δH(t)
f |Ψ̃(t)i (15.8)
and X
|Ψ̃(t)i = λk |Ψ̃(k) (t)i . (15.9)
k

For this section, we assume the perturbation is turned on at t = 0. So |Ψ̃(0) i is


time-independent, and |Ψ̃(n) (t = 0)i = 0 for n ≥ 1.
Substitution into the differential equation gives
X X
ih̄ λk ∂t |Ψ̃(k) i = λk δH
f |Ψ̃(k) i
k k

whence power matching gives:

ih̄∂t |Ψ̃(0) (t)i = 0


ih̄∂t |Ψ̃(1) (t)i = δH
f |Ψ̃(0) (t)i

ih̄∂t |Ψ̃(2) (t)i = δH


f |Ψ̃(1) (t)i

ih̄∂t |Ψ̃(3) (t)i = δH


f |Ψ̃(2) (t)i

and so forth. To find the first correction, we can just integrate:


Z t
1 f 0 ) |Ψ(0)i dt0
|Ψ̃(1) (t)i = δH(t (15.10)
ih̄ 0

45
Tony Zhang 16 March 13, 2017 (recitation)

16 March 13, 2017 (recitation)


Example: perturbative solution to δ-function perturbation, which we considered
in section 15.3.1.
First, we note that the nth order correction is really
Z Z t n
(n) 1 o
|Ψ̃ (t)i = · · · T δH(t
f 1 )δH(t
f 2 ) . . . δH(t
f n ) dt1 dt2 . . . dtn
n! (ih̄)n 0
(16.1)
where we have performed a time-ordering.
Blah blah blah, doing this to second order gives us the correct result to
second order.

16.1 Instantaneous approximation


This class of problems is as follows: our Hamiltonian shifts quickly from time-
independent H0 to time-independent H1 :
(
H0 t < − 21 ε
H(t) = .
H1 t > 12 ε

Let the transition time be ∆t = ε → 0. If it’s “short enough”, the state will
have no time to change, and will remain the same. Let’s figure out what “short
enough” is.
As always, we have the time-dependent Schrödinger equation:
∂ |ψi
ih̄ = H(t)ψ(t).
∂t
Integrating,
     Z 1 ε
1 1 2
ih̄ ψ ε −ψ − ε = H(t0 )ψ(t0 ) dt0 ∼ ∆E ∆T ψ(0).
2 2 − 21 ε

We want the left hand side to be basically 0, so the condition for our approxi-
mation holding is for ∆E ∆t  h̄.
One can derive this relation more rigorously by actually finding an expression
for the energy uncertainty. But the intuition is simple: the relative phase
differences in quantum state time evolution grow as exp( i∆E t
h̄ ).

16.1.1 Example: β-decay

In β-decay, we go from a neutron atom with nuclear charge Ze to a singly


positive ion of charge (Z + 1)e with a relativistic electron ejected. (A neutron
turned into a proton, ejecting a fast electron in the process.)
The speed at which the electron leaves is fast enough that the sudden
approximation holds: we can model the system as jumping from the Ze nucleus
to the (Z + 1)e one.
For example, in the β-decay of tritium:
3H −−→ 3He+ + e− ,

46
Tony Zhang 16 March 13, 2017 (recitation)

the ejected electron has average energy 5.7 keV, which translates to a velocity
of v = 0.15c. So the electron leaves within ∆t = av0 = 1.2 × 10−18 s. The
typical energy in a hydrogen atom is 10 eV, so we have ∆Eh̄
= 6.4 × 10−17 s, so
∆E∆t  h̄, and we can use the instantaneous approximation.
If the tritium atom began in the ground state, we could figure out the
time-evolution of the helium ion (another hydrogenic system) by assuming the
helium ion starts in the hydrogen ground state
1 r
ψ100 (r) = p e− a0 ,
πa30

which we would decompose into helium ion states, which look like the hydrogen
ones, with a0 7→ 12 a0 . We’d evolve each component separately, as usual.

47
Tony Zhang 17 March 15, 2017 (recitation-cum-lecture)

17 March 15, 2017 (recitation-cum-lecture)


17.1 Time-dependent perturbation theory: review
We’ll finish up some time-dependent perturbation theory stuff. Recall we were
treating time-dependent Hamiltonian
H(t) = H (0) + δH(t) (17.1)
where H (0) is time-independent and known. We defined
iH (0) t
|Ψ̃(t)i = e h̄ |Ψ(t)i (17.2)
which is much easier to deal with.
We found a Schrödinger equation for |Ψ̃(t)i:

∂ |Ψ̃i f |Ψ̃i
ih̄ = δH (17.3)
∂t
where (0) t iH (0) t
f = e iHh̄
δH δHe− h̄ . (17.4)

Expanding in H (0) eigenstates,


X
|Ψ̃(t)i = cn (t) |ni , (17.5)
n

where the coefficient functions obey


X
ih̄ċm (t) = eiωmn t δHmn (t)cn (t) (17.6)
n

with
Em − En
ωmn = . (17.7)

Finally, considering a perturbative expansion
|Ψ̃(t)i = |Ψ̃(0)i + |Ψ̃(1) (t)i + . . . , (17.8)
we find the first order correction is
Z t f 0)
δH(t
|Ψ̃(1) (t)i = |Ψ(0)i dt0 . (17.9)
0 ih̄

17.2 Transition probability


A common question to ask: if |Ψ(0)i = |ni, what is the probability of transi-
tioning into state m 6= n? To first order:
Pr (t) = |hm|Ψ(t)i|2
m←n
iH (0) t
= |hm|e− h̄ |Ψ̃(t)i|2
= |hm|Ψ̃(t)i|2
  2
= hm| |Ψ̃(0)i + |Ψ̃(1) (t)i + O(λ2 )

= |hm|Ψ̃(1) (t)i|2

48
Tony Zhang 17 March 15, 2017 (recitation-cum-lecture)

But then we have


Z 2 Z
t hm|δH(t
f 0 )|ni t iω t0 δHmn (t0 ) 0 2

0
Pr (t) = dt = e mn dt (17.10)

m←n 0 ih̄ 0 ih̄

So we’re taking a “Fourier transform” in some sense.


Remark 17.1. Many authors more intuitively write Prn→m . We reverse the
order so the indices match up nicely.

17.3 First-order coefficients


Now let’s consider a perturbative expansion of the cn . Then
X
|Ψ̃(t)i = cn (t) |ni
n
X X
= c(0)
n |ni + λc(1)
n (t) |ni + . . .
n n
| {z } | {z }
|Ψ̃(0)i λ|Ψ̃(1) (t)i

So the first order perturbations to the coefficients are

c(1)
m (t) = hm|Ψ̃
(1)
(t)i
* Z +
t δH(tf 0) X
= m c(0)
n n

0 ih̄ n

X tZ
0 δHmn
= eiωmn t c(0) dt0 (17.11)
n 0 ih̄ n

17.4 Fermi’s Golden Rule with constant transitions: intro


Fermi’s Golden Rule is a remarkable result that lets us calculate very nontrivial
transition amplitudes.
Joke. Fermi’s Golden Rule was actually probably created by Dirac.

Suppose the Hamiltonian experiences just one discontinuous shift:

H(t) = H (0) + V (17.12)

where V is constant in time turned on for t ≥ 0. Note that V is still an operator,


and not necessarily constant on all states. It’s just time-independent.
Let i 6= f denote initial and final energy eigenstates. We would like to find
the transition probability f ← i. Then for some time t0 after the perturbation
is turned on,
Z t0
(1) 0 Vf i
cf (t0 ) = eiωf i t dt0
0 ih̄
Vf i
1 − eiωf i t0

=
Ef − Ei
1
Vf i e 2 iωf i t0 −2i sin 21 ωf i t0

=
Ef − Ei

49
Tony Zhang 17 March 15, 2017 (recitation-cum-lecture)

Then to first order,

4|Vf i |2 sin2 21 ωf i t0
2 
(1)
Pr (t0 ) = cf (t0 ) = . (17.13)

f ←i (Ef − Ei )2

This expression is interesting.

• We have oscillatory behavior!


• If the oscillation amplitude is too large, we expect to need second-order
corrections.

• In the Ef → Ei limit, it seems we would get a divergence because of the


Ef − Ei in the denominator. But actually, it’s fine, because in this limit
the sine is also small:
|Vf i |2 t20
Pr (t0 ) → . (17.14)
f ←i h̄2
This approximation is valid when

ωf i t0  1.

That is, a smaller energy split makes this result valid for longer.

Now consider what happens if our spectrum is continuous (e.g. the scattering
states of hydrogen). It turns out we’ll be able to apply the same logic to such
states as well; transition probabilities are not transition probability densities,
etc.

17.4.1 Example: Auto-ionization of helium

Let’s consider helium (Z = 2, with two electrons). Suppose that the two
electrons are in hydrogenic states with principal quantum numbers n1 , n2 . The
total energy of the system is then
   
2 1 1 1 1
En1 ,n2 = −(13.6 eV )Z + = −54.4 eV + (17.15)
| {z } n21 n22 n21 n22
Ry

The first few energies are

n1 n2 En1 ,n2 /eV


1 1 −108.8
1 2 −68.0
2 2 −27.2

So it seems like we have a discrete spectrum until we get to energy zero. But
that’s not true. Indeed, we have E∞,1 = −54 eV; therefore, our continuum
actually starts here. If we’re in a bound state, like E2,2 , with greater energy,
there is actually a probability we transition into a state with a free electron;
this is the Auger effect.

50
Tony Zhang 18 March 16, 2017

18 March 16, 2017


18.1 Fermi’s Golden Rule with constant transitions
Suppose we have time-dependent Hamiltonian H that jumps from one constant
Hamiltonian H (0) to H (0) + V for some V constant in time.
(
H (0) t≤0
H= (18.1)
H (0) + V t ≥ 0

Then we have first-order transition probabilities


21

2 4 sin 2 ωf i t0
Pr (t0 ) = |Vf i | , (18.2)
f ←i (Ef − Ei )2
which is oscillatory, but also quadratic in the ωf i t0  1 limit.
It’s straightforward to sum these probabilities when we have a discrete
spectrum, as we might want to do if we want to find the probability of a class
of transitions. But what if we want to sum over the continuum? For example,
we might want to compute the probability of ionization of hydrogen after some
discontinuous shift in the Hamiltonian.
Formally, we need a measure of integration on the continuous part of the
spectrum of H (0) . To do so, we’ll impose some regularization to discretize the
spectrum and take a limit.
Suppose our universe is a huge box of side length L with periodic boundary
conditions. Then for a normalized plane wave
1 ~
ψ(x) = √ eik·~x , (18.3)
L 3

we must have
L~
k = (nx , ny , nz )

be an integer vector. The number of states within a volume element in ~n space
is thus  3
L
∆N = dnx dny dnz = d3 k.

Now imagine ~k-space in spherical coordinates:

d3 k = k 2 dk dΩ

where dΩ = sin θ dθ dφ is a solid angle element. But recall that plane wave
energies are given by
h̄2 k 2
E= ,
2m
so that
km
d3 k = 2 dE dΩ

and  3
L m
∆N = k dΩ dE
2π h̄2
| {z }
ρ(E)

51
Tony Zhang 18 March 16, 2017

0.8

0.6

0.4

0.2

0
−4π −3π −2π −π 0 π 2π 3π 4π

2
Figure 1: The square of the sinc function sinc2 z = sinz2 z . The central lobe
captures 90% of the mass. Within 2π of the origin lies 95% of the mass; within
10π, 99%.

where we’ve defined a state density per unit energy in solid angle dΩ.
Therefore, the probability of transitioning into an energy eigenstate with
energy Ef ∈ F is Z
Pr (t0 ) = Pr (t0 )ρ(Ef ) dEf .
F ←i F f ←i

Question 18.1. The density blows up as we decrease regulation: ρ(Ef ) ∼ O(L3 ).


How does our integral remain finite?

Answer. The transition probabilities will correct for the explosion!


|Vf i |2 = |hf |V |ii|2 ∼ |ψ|2 = O(L−3 ).

Substituting in first-order transition probabilities (Equation 18.2) and as-


suming F is actually some tiny interval Ei ± ∆ of the spectrum centered at Ei ,
we can safely assume |Vf i |2 and ρ(E) are constant:
2 1

2 sin 2 ωf i t 0
Z
Pr (t0 ) = 4 |Vf i | ρ(Ef ) dEf
F ←i F (Ef − Ei )2
Z Ei +∆ 2 2 1 
4|Vf i |2 h̄ sin 2 ωf i t0
= ρ(Ei ) dEf (18.4)
h̄2 Ei −∆ (Ef − Ei )2

Notice the sinc squared. Looking at Figure 1, the main contribution to such
an integral comes from the central lobe of the sinc squared. Indeed, we will
capture 90% of the integral of sinc2 z in the first lobe |z| < π.
We’re in the first lobe when |ωf i t0 | < 2π, or equivalently,
2πh̄ 2πh̄
Ei − < Ef < Ei + .
t0 t0
Changing to integration variable
1
u= ωf i t0 ,
2
52
Tony Zhang 18 March 16, 2017

the integral becomes


∆t0 ∆t0
2h̄ sin2 u sin2 u
Z Z
2h̄ 1 2h̄
I≈ 2 = h̄t0 du.

∆t0
2h̄
t0 4u t2
2 −
∆t0
2h̄
u2
0

∆t0
If 2h̄  π, this integral is basically the full integral over all u:
Z ∞
1 1
I ≈ h̄t0 sinc2 u du = πh̄t0 .
2 −∞ 2

Plugging back into Equation 18.4,



Pr (t) = |Vf i |2 ρ(Ef = Ei )t (18.5)
F ←i h̄
whence we get a transition rate W = ∂t PrF ←i (t):


W = |Vf i |2 ρ(Ef = Ei ) . (18.6)

This transition rate tells us the probability per unit time that we transition into
a state with energy Ei ± ∆ where ∆  2πh̄ t0 .
3

What is the condition for this approximation to hold? We want |Vf i |


sufficiently large such that ρ(E) is about constant.
Also, since we have a probability rate W , we need W t  1.

18.2 Harmonic perturbations: introduction


Consider a harmonic perturbation

H(t) = H (0) + δH(t) (18.7)

with 
0
 t<0
0
δH = 2H cos ωt 0 < t < t0 . (18.8)

0 t > t0

where H 0 is time-independent and ω > 0. The factor of 2 is there for convenience.


(The Golden Rule will look the same!)
Recall that
XZ t
0 δH (0) 0
c(1)
m (t) = eiωmn t c dt
n 0 ih̄ n
so that Z t0
(1) 1 0
cf (t, t0 ) = eiωf i t 2Hf0 i cos(ωt0 ) dt0 (18.9)
ih̄ 0
which is time-independent. Therefore,
Hf0 i ei(ωf i +ω)t0 − 1 ei(ωf i −ω)t0 −1
 
cf (t0 ) = − + . (18.10)
h̄ ωf i + ω ωf i − ω
3 Well, it actually gives the probability of transitioning into any continuum state, but

the likelihood of going far from Ei is negligible, due to the sinc2 u supression. Assuming a
transition occurs, about 90% of the time, we’ll be within 2πh̄
t
of Ei .
0

53
Tony Zhang 18 March 16, 2017

Which of these terms is dominant depends on the denominators, which in turn


depend on the energy difference Ef − Ei . If it’s about −h̄ω, the second term
denominates; +h̄ω, the first.
Note: we will never be interested in a case where both terms contribute
significantly.

54
Tony Zhang 19 March 21, 2017

19 March 21, 2017


19.1 Fermi’s Golden Rule with harmonic perturbation
Recall that a harmonic perturbation involves a perturbing term

0
 t<0
δH(t) = 2H 0 cos ωt 0 < t < t0

0 t > t0

After this perturbation, the probability amplitude of being in an energy


eigenstate f is, to first order,
Hf0 i ei(ωf i +ω)t0 − 1 ei(ωf i −ω)t0 − 1
 
cf (t0 ) = − + ,
h̄ ωf i + ω ωf i − ω
as we found last time. We focus on the second term, corresponding to absorption.
That is, suppose |ωf i − ω|  |ωf i + ω|, so that the second term dominates the
first.
Then we can approximate
Hf0 i ei(ωf i −ω)t0 − 1
cf (t0 ) = −
h̄ ωf i − ω
0
Hf i 1 i(ω −ω)t0 2i sin 12 i(ωf i − ω)t0

=− e2 f i
h̄ ωf i − ω
0 2 2 1

2
|H |
f i 4 sin 2 (ωf i − ω)t0
|cf (t0 )| =
h̄2 (ωf i − ω)2

What follows will look much like the constant transitions case we worked out.
Suppose we’re interested in the probability of transition into any state within
some small ∆ of Ei + h̄ω. So we integrate, making the same approximations as
in the constant transitions case:
|Hf0 i |2 4 sin2 21 (ωf i − ω)t0
Z Ei +h̄ω+∆ 
Pr = ρ(Ei + h̄ω) dEf
f ←i h̄2 Ei +h̄ω−∆ (ωf i − ω)2
Z ∆t
|Hf0 i |2
0
2h̄ 4 sin2 u 2h̄
= 2 ρ(Ei + h̄ω) du
h̄ ∆t
− 2h̄0 u2 t42 t0
0
∆t0
2|Hf0 i |2
Z 2h̄
= t0 ρ(Ei + h̄ω) sinc2 u du
h̄ −
∆t0
2h̄

Taking the limits of the integral to infinity and dividing out t0 , we get our
transition rate
2π 0 2
Γf ←i = |H | ρ(Ei + h̄ω) (19.1)
h̄ f i

Again, this rate is valid for times


2πh̄ 1
 t0 
∆ Γ
where ∆ roughly gives our desired energy uncertainty.
An identical result holds for

55
Tony Zhang 19 March 21, 2017

19.2 Example: Ionization of hydrogen


Let’s look at the ionization of hydrogen. What happens when we shine some
light on a ground state hydrogen atom? How much energy does it take to eject
the electron, and in what direction will the electron fly away?
Consider a traveling EM wave producing a perturbation

δH = −eΦ(t)

where Φ is the electric potential produced from the electric field


~
E(t) = E(t)ẑ = 2E0 cos ωtẑ.

Assuming electrostatics, the potential is

Φ = −E(t)z.

Therefore,
δH = 2 eE0 r cos θ cos ωt.
| {z }
H0

Assume the hydrogen atom starts in the ground state:


1 r
ψi (r) = p e− a0
πa30

and post-ejection, the electron is in a plane wave:


1 ~
u~k = √ eik·~r
L3
where L is our regularization term from before.
Now switch coordinates, so that we’re still in spherical coordinates, but with
~k now pointing along ẑ. Then let the angular components of E ~ and ~r be (θ, φ)
0 0 ~ and ~r be θ00 .
and (θ , φ ), respectively. Also, let the angle between E
To use the Golden Rule, we need the
~
e−ik·~r
Z
1 r
hf |H 0 |ii = d3 x 3 eE0 r cos θ00 p e− a0
R3 L 2 πa30
Z
eE0 0 r
=p r2 dr sin θ0 dθ0 dφ0 e−ikr cos θ r cos θ00 e− a0
πL3 a30

This integral is not so easy. For example, we normally don’t have φ0 depen-
dence; however, there’s a stealthy dependence here in θ00 . Let’s handle that
term. We have
~ · r̂
|E| cos θ00 = E
= Ex nx + Ey ny + Ez nz
= |E| (sin θ cos φ sin θ0 cos φ0 + sin θ sin φ sin θ0 sin φ0 + cos θ cos θ0 )
cos θ = sin θ sin θ0 cos(φ − φ0 ) + cos θ cos θ0
00

56
Tony Zhang 19 March 21, 2017

Now the cos(φ − φ0 ) term will vanish when we integrate in φ0 , so we can


ignore that term. Therefore, we continue:
Z
eE0 0 r
0
hf |H |ii = p 3
r3 dr sin θ0 dθ0 dφ0 e−ikr cos θ cos θ cos θ0 e− a0
3
πL a0
Z ∞ Z 1
eE0 − ar 0
=p 3
2π cos θ 3
r dr e 0 d(cos θ0 )e−ikr cos θ cos θ0
3
πL a0 0 −1

Already, the angular dependence pops out. It’s not that surprising: given an
electric field, we expect ejection probability to be concentrated in the direction
of the electric field, with no probability perpendicular.
Omitting the remaining steps of integration (which are now really straight-
forward), we get our desired overlap:
√ ka4 1
hf |H 0 |ii = −32i π(eE0 a) p 0 3 2 a2 )3
cos θ. (19.2)
3
L a0 (1 + k 0

Then
k 2 a50 1
|Hf i |2 = 1024π(eE0 a)2 cos2 θ
L (1 + k 2 a20 )6
3

Now we only need the state density to use Fermi’s Golden Rule:
L3 mh̄2
ρ(E) = sin θ dθ dφ (19.3)
8π 3 k | {z }
dΩ

where E appears implicitly through the relation


h̄2 k 2
= Ef . (19.4)
2m
Then the Golden Rule gives transition probability rate

W = ρ(Ef )|Hf0 i |2

or
256 (eE0 a0 )2 ma20 k 3 a30
W = cos2 θ dΩ (19.5)
π h̄ 2
h̄ (1 + k 2 a20 )6

Finally, we might integrate this rate over the entire sphere to get an ejection
rate. The angular dependence can be integrated as
Z
4
dΩ cos2 θ = π
3
whence
1024 (eE0 a0 )2 ma20 k 3 a30
W = . (19.6)
3 h̄ h̄2 (1 + k 2 a20 )6
Remark 19.1. Two nice integrals to remember:
Z Z
1 1 1 2
2
dΩ cos θ = dΩ sin2 θ = .
4π 3 4π 3
The way to remember: cos2 θ is concentrated at the poles and sin2 θ is concen-
trated around the much larger equatorial region.

57
Tony Zhang 19 March 21, 2017

19.3 Light and atoms


Consider an atom with nuclear charge Z and energy levels Ea < Eb . Define the
characteristic frequency of these two levels:
Eb − Ea
ωba = .

Suppose we shine light of such frequency on the atom. There are two things
that could happen:

1. If the electron is in |ai, the photon can be absorbed with some absorption
rate, exciting the electron to |bi.
2. The electron starts in |bi, and we get stimulated emission of a photon.
The electron decays to |ai; one photon goes in, two go out. (Lasers work
like this: you excite a bunch of atoms, then shine a little light on them.
That light gets amplified.)

We already have the tools to analyze these two transitions (i.e. Fermi’s Golden
Rule).
But there’s another transition that’s super important: spontaneous emis-
sion. Spontaneous emission is hard to explain at first, because shouldn’t energy
eigenstates be stationary? The way to think about it is that there are vacuum
fluctuations of the EM field that perturb the system ever so slightly.

58
Tony Zhang 20 March 22, 2017 (recitation)

20 March 22, 2017 (recitation)


20.1 Scattering and the Born approximation
We apply Fermi’s Golden Rule to the problem of scattering. Suppose we have a
Hamiltonian
p2
H= + Ṽ (x).
2m
3 ~ 3 ~
There is an incident plane wave L− 2 eiki ·~x and scattered wave L− 2 eikf ·~x for a
box size L. The energies are given by

h̄2 |~k|2
E= .
2m

Recall that the state density is

L3 km
ρ(E) = dΩ (20.1)
(2π)3 h̄2
whence Fermi’s Golden Rule

Γf ←i = ρ(Ef )|hf |δH|ii|2 (20.2)

gives differential transition rate
kf m
dΓ = |Ṽ (~kf − ~ki )|2 dΩ
(2π)2 h̄3 L3

where Ṽ is the three-dimensional Fourier transform of V .


Welp. There’s a L3 in there, which seems unphysical. But actually, it’s quite
physical. Since we assumed a plane wave, for larger box size, the plane wave
(basically uniform over the entire box) is more and more unlikely to hit the
scattering potential.
How do we extract a physically meaningful quantity out then? We want the
scattering cross-section:
rate of particles scattering into dΩ dΓ
dσ = = (20.3)
incoming particles per unit time per unit area Φ
The cross-section is an interesting quantity because in the classical picture

where we scatter off of a solid particle the integral of dΩ (the ratio between
the total number of particles scattered and the particle flux density) is just the
cross-sectional area of the particle.
We find the flux Φ. Since the incoming particles come in plane waves, they
~
approach with fixed velocity ~v = h̄mk . The particle density in some slab of area
L2 and thickness v dt
dN = v dtL2 ρparticles .
But
1
ρparticles =
L3
so that
dN v
Φ= = 3.
L2 dt L
59
Tony Zhang 20 March 22, 2017 (recitation)

and our cross section is


2
m
~ ~

dσ = 2 Ṽ (kf − ki ) dΩ. (20.4)

2πh̄
This is a first order Born approximation. Approximation, because we only
went to first order in time-dependent perturbation theory.

20.2 About probabilities


The Γ from Fermi’s Golden Rule is a probability rate. By conservation of
probability, the probability of not making the transition is 1 − Γf ←i t. But
probabilities can’t be negative, so evidently we need t  O(Γ−1 ) (which is
almost always true).

Recall we also need t  ∆ for Fermi’s Golden Rule. So we have this double
constraint on time t. In reality, only this second one makes a difference.

60
Tony Zhang 21 Apri 4, 2017

21 Apri 4, 2017
We continue our discussion of light and atoms. Today, we will cover Einstein’s
discovery of spontaneous emission and the argument he used to calculate its
rate. Note that this argument is thermodynamic, and does not come from first
principles (i.e. not derived by interpreting spontaneous emission as stimulated
emission by vacuum fluctuations).

21.1 Einstein’s argument for spontaneous emission


Consider two atomic energy levels Ea < Eb with associated transition frequency
ωba = Eb −Eh̄
a
. Suppose we have a large ensemble of such atoms in thermal
1
equilibrium with a thermal bath of radiation at temperature T . Define β = kT
where k is Boltzmann’s constant. Suppose Na and Nb of them are in state |ai
or |bi, respectively. Then enforcing equilibrium:

1. Ṅa = Ṅb = 0
Nb e−βEb
2. Na = e−βEa
= e−βh̄ωba .
3. Assuming thermal radiation, the energy density in frequency interval dω is
 3 
h̄ ω dω
U (ω) dω = 2 3 . (21.1)
π c eβh̄ω − 1

What transition mechanisms do we know so far?

Process Transition Transition rate


Absorption |ai → |bi Bab U (ωba )Na
Stimulated emission |bi → |ai Bba U (ωba )Nb

where we’ve considered some coefficients Bab and Bba that give the transition
rate.
Then we have

0 = Ṅb = −(rate of stimulated emission) + (rate of absorption)


= −Bba U (ωba )Nb + Bab U (ωba )Na
= (Bab Na − Bba Nb )U (ωba )
= Na (Bab − Bba e−βh̄ωba )U (ωba )

But Bab and Bba are atomic properties, and thus independent of temperature
T . So none of the three factors is zero in general, and we get a contradiction.
This argument is due to Einstein; when he saw this contradiction, he added
a third process: spontaneous emission. We summarize our three pathways
in Table 1.

61
Tony Zhang 21 Apri 4, 2017

Table 1: Transition processes in atoms between states of energy Ea < Eb .

Process Transition Transition rate


Absorption |ai → |bi Bab U (ωba )Na
Stimulated emission |bi → |ai Bba U (ωba )Nb
Spontaneous emission |bi → |ai ANb

Now we try our above argument again, this time with spontaneous emission:

0 = Ṅb = −ANb − Bba U (ωba )Nb + Bab U (ωba )Na


 
Na
A = Bab − Bba U (ωba )
Nb
 −1
A Na Bba
U (ωba ) = −
Bab Nb Bab
3
h̄ωba 1 A 1
= Bba
π 2 c3 eβh̄ωba − 1 Bab eβh̄ωba − B
ab

whence
Bab = Bba (21.2)
and
A h̄ω 3
= 2 ba3 . (21.3)
Bab π c
Recall that in first-order time-dependent perturbation theory, the symmetric
transition rates are equal. We see that here too. Also, we have a nice way to
get Einstein’s A coefficient, the rate of spontaneous emission.

Remark 21.1. Some people like to say that all emission is stimulated emission.
It’s just that there are vacuum fluctuations in the electromagnetic field that
can stimulate emission, giving what we like to call spontaneous emission. But
it’s inconvenient to actually do calculations about spontaneous emission by
considering vacuum fluctuations and such.
So in the end, paraphrasing Griffiths, spontaneous emission is just stimulated
emission from photons we didn’t put there.

21.2 Atom-light interaction


We will focus on the effect of the electric field only. Indeed, the magnetic field
effect will be a factor vc ∼ α ≈ 137
1
smaller.
Now the characteristic wavelength λ of typical atomic transitions is around
4000 Å - 8000 Å, much larger than the atom itself. Therefore, we neglect the
spatial variation of the electric field:

E(t) = 2E0~n cos ωt

so that we get potential


~
Φ(~r, t) = −~r · E(t)
and perturbation
~
δH = qΦ(~r, t) = − q~r ·E(t)
|{z}
d~

62
Tony Zhang 21 Apri 4, 2017

where we’ve found the dipole moment of the atom. Finally, we have

δH = 2(−d~ · ~nE0 ) cos ωt.


| {z }
H0

This harmonic perturbation gives transition probabilities


0 2 sin2 1 (ω

4|Hab | 2 ba − ω)t
Pr (t) = Pr (t) =
b←a a←b h̄2 (ωba − ω)2
4E02 |(d~ · ~n)ab |2 sin2 12 (ωba − ω)t

=
h̄2 (ωba − ω)2

Now
|E|2 E2
UE = = 0 cos2 ωt
8π 2π
whence a time average gives
1 2
hUE it = E .
4π 0
The total time-averaged energy of the electromagnetic field (including the
magnetic field) is then double that:
1 2
hU i = E .
2π 0
Therefore, Z Z
E02 (ω) dω = 2π U (ω) dω.

We’ll use this result right now in our computation of the transition probability.
Now in a thermal bath, there are multiple incoherent perturbations of
differing ω. Incoherence means probabilities add classically:
2 1
|(d~ · ~n)ab |2

2 sin 2 (ωba − ω)t
Z
Pr (t) = 4E0 dω
a←b h̄2 (ωba − ω)2
|(d~ · ~n)ab |2 sin2 21 (ωba − ω)t
Z 
= 2π U (ω) dω
h̄2 ωba −ω 2

2

Now we’re in Fermi’s-Golden-Rule land! Again, the sinc is peaked for large t, so
pulling out U (ω) and making a substitution x = 12 (ω − ωba )t, we recognize our
Dirichlet integral:

|(d~ · ~n)ab |2 sin2 12 (ωba − ω)t


Z 
Pr (t) = 2π U (ωba ) dω
h̄2 ωba −ω 2

a←b
2
| {z }
2πt

4π |d~ab · ~n|2
2
= U (ωba )t
h̄2
and get our transition rate:

4π 2 |d~ab · ~n|2
Ra←b = U (ωba ) . (21.4)
h̄2

63
Tony Zhang 21 Apri 4, 2017

The prefactor on U (ωba ) is the coefficient Bab .


Now in an actual thermal bath, the polarization ~n is random. So we really
want to average over ~n on the unit sphere:
* 2 +
D E X
~ 2
|dab · ~n| =
i
dab ni


i
* ! +
X X j
= diab ni  (dab )∗ ni 
i j
X
= diab (djab )∗ hni nj i
ij
1
= |d~ab |2
3
where we use the fact that hni nj i = 31 δij . Then averaged,

4π 2 ~ 2
Ra←b = |dab | U (ωba ) . (21.5)
3h̄2

Then we recognize
4π 2 ~ 2
Bab = Bba = |dab | . (21.6)
3h̄2
h̄ω 3
and A = Bab π2 cba3 gives
3
4 ωba
A= |d~ab |2 . (21.7)
3 h̄c3
1
whence we get lifetime τ = A and

Nb (t) = Nb (0)e−t/τ .

Real atoms have multiple relaxation pathways from any given excited state.
The sizes of the various A coefficients thus determine the selection rules, which
tell us which atomic transitions are actually possible in a reasonable amount of
time.
Also, this sort of argument shows that almost any particle, any quantum
system, is unstable in some way. Even protons and neutrons! But some transition
pathways are rather forbidden in the sense that their A coefficients are small.

Question 21.2. On the other hand, electrons are actually stable. Why?

Answer. They don’t have a decay pathway available to them— they’re the
least massive known negatively charged particles.

Last thing: in general, we can find the lifetime of a state by inverting the
total spontaneous emission rate: the sum of all the relevant A coeficients.

64
Tony Zhang 22 April 5, 2017 (recitation)

22 April 5, 2017 (recitation)


22.1 Fermi’s Golden Rule TODO TODO TDOO TODOT ODO
TODO
something about Fermi’s Golden Rule and Cauchy distributions
use section 3.2 of https://www.physics.drexel.edu/~bob/Manuscripts/
time_dep_pt.pdf

22.2 Selection rules


Recall that the spontaneous decay of an excited state of energy Eb to state Ea
has rate given by Einstein’s A coefficient:

4 αω 3
A= |hb|~x|ai|2 . (22.1)
3 c2
Now it turns out that A will often be 0 because the matrix element there
vanishes: some transitions are rather forbidden. It would be nice to know if it
does before trying to bash out the integral.
Suppose we’re in hydrogen (or some other rotationally invariant system),
with states |n`mi (ignoring fine structure, spin-orbit coupling, and such). What
sorts of decay are possible?

Proposition 22.1 (Selection rules)


hn0 `0 m0 |~x|n`mi = 0 unless |m0 − m| ≤ 1 and |`0 − l| = ±1.

Proof. Recall Li = εilm xl pm so that

[Li , xj ] = εilm [xl pm , xj ] = εilm (xl [pm , xj ]) = −ih̄εilm xl = ih̄εijk xk .

As a result, [Lz , x] = ih̄y, [Lz , y] = −ih̄x, and [Lz , z] = 0. Then

0 = hn0 `0 m0 |[Lz , z]|n`mi


= h̄(m0 − m) hn0 `0 m0 |z|n`mi

so either m0 = m or hn0 `0 m0 |z|n`mi = 0.


Also,

ih̄ hn0 `0 m0 |y|n`mi = hn0 `0 m0 |[Lz , x]|n`mi


= h̄(m0 − m) hn0 `0 m0 |x|n`mi

Finally, exploit the last unused commutation:

−ih̄ hn0 `0 m0 |x|n`mi = hn0 `0 m0 |[Lz , y]|n`mi


= h̄(m0 − m) hn0 `0 m0 |y|n`mi

These two relations between the x sandwich and the y sandwich force upon us
hn0 `0 m0 |x|n`mi = (m0 − m)2 hn0 `0 m0 |x|n`mi. So either m0 − m = ±1 or both
sandwiches vanish.

65
Tony Zhang 22 April 5, 2017 (recitation)

Therefore, all three sandwiches vanish unless m0 − m = 1, 0, −1, proving the


first part. This fact isn’t that surprising, and we could have just bashed out an
integral. Indeed, it boils down to evaluating
Z
Y`∗0 m0 (Ω)(sin θ cos φ, sin θ sin φ, cos θ)Ylm (Ω) dΩ

and using the fact that Ylm (θ, φ) = f (θ)eimφ .


For the second selection rule, we will use the commutator

[L2 , [L2 , xi ]] = 2h̄2 (xi L2 + L2 xi ). (22.2)

Proof sketch. Using [Li , xj ] from before, find that [L2 , xi ] = 2ih̄(εijk xj Lk −ih̄xi ).
Then use xk Lk = 0. 

Sandwiching the left hand side of the commutation relation,

hn0 `0 m0 |[L2 , [L2 , xi ]]|n`mi = h̄2 (`0 (`0 + 1) − `(` + 1)) hn0 `0 m0 |[L2 , xi ]|n`mi
= h̄4 (`0 (`0 + 1) − `(` + 1))2 hn0 `0 m0 |xi |n`mi

Doing the same to the right hand side,

2h̄2 hn0 `0 m0 |(xi L2 + L2 xi )|n`mi = 2h̄4 (`0 (`0 + 1) + `(` + 1)) hn0 `0 m0 |xi |n`mi
= 2h̄4 (`0 (`0 + 1) + `(` + 1)) hn0 `0 m0 |xi |n`mi

Equating the two, either the xi matrix element vanishes, or

(`0 (`0 + 1) − `(` + 1))2 − 2`0 (`0 + 1) − 2`(` + 1) = 0.

It remains to factor this expression. The term in the square is


2
`0 (`0 + 1) − `(` + 1) = `0 − `2 + `0 − `
= (`0 − l)(`0 + l) + `0 − `
= (`0 − l)(`0 + ` + 1).

The original polynomial is then, upon factoring the second half,

0 = (`0 − `)2 (`0 + ` + 1)2 − ((`0 + ` + 1)2 + (`0 − `)2 − 1)


= ((`0 − `)2 − 1)((`0 + ` + 1)2 − 1)

by Simon’s Favorite Factoring Trick.


Therefore, either `0 − ` = ±1 or `0 + ` + 1 = ±1. But this latter case gives
` + ` = 0, in which case the matrix element hn0 `0 m0 |~x|n`mi = 0 as well.
0

Example 22.2 (Decay of |4, 1, −1i)


Suppose we start in |4, 1, −1i (a 4p state). The selection rules tell us we
can only decay into s or d states, and m = −2, −1, 0. The only such states
are the 1s, 2s, 3s states and the three 4d states with valid m quantum
number.

Problem 22.3. Show that the total decay rate of a state |n`mi is independent
of m.

66
Tony Zhang

Part V

The adiabatic approximation

67
Tony Zhang 23 April 6, 2017

23 April 6, 2017
Today, we begin our unit on the adiabatic approximation.

23.1 Classical adiabatic change


Today we’ll talk about the adiabatic approximation, an approximation we
can make when H changes very slowly in a precise sense: when it changes
adiabatically. The prototypical adiabatic change in classical mechanics is that
of a pendulum whose length is slowly decreased so as to change oscillation
frequency ω. The classical Hamiltonian looks like

p2 1
H(p, x; t) = + mω 2 (t)x2 . (23.1)
2m 2

Recall Hamilton’s equations of motion:


∂H ∂H
ẋ = ṗ = − (23.2)
∂p ∂x
which give us
dH ∂H ∂H ∂H ∂H
= ṗ + ẋ + = (23.3)
dt ∂p ∂x ∂t ∂t
Remark 23.1. In quantum mechanics, there are analogues to these equations
of motion. Indeed, we can interpret p = −ih̄∂x , so that
 
d ∂H
ih̄ hpi = h[p, H]i = −ih̄ .
dt ∂p
A similar computation with x = ih̄∂p gives
   
d ∂H d ∂H
hxi = hpi = − (23.4)
dt ∂p dt ∂x

Back to our classical oscillator. For that Hamiltonian,


dH ∂H
= = mω(t)ω̇(t)x2 . (23.5)
dt ∂t

Now we ask: what is an adiabatic change? In our harmonic oscillator


example, if we let ω change smoothly over time τ , our change is adiabatic if

τ  T = ω(t) , where T is the instantaneous period of oscillation. But this
bound should leave you unsatisfied: it only defines an adiabatic change for this
example.
What sort of bound can we derive in any slowly changing Hamiltonian?

Proposition 23.2
The quantity
H(t)
I(t) = (23.6)
ω(t)
is an adiabatic invariant for the classical harmonic oscillator system.

68
Tony Zhang 23 April 6, 2017

We’ll see what we mean by an adiabatic invariant.

Proof. Compute:

dI ω(t) dH
dt − H dt

=
dt ω2
∂H
ω − H ω̇
= ∂t 2
ω
p2
mω 2 ω̇x2 − ( 2m + 21 mω 2 x2 )ω̇
=
ω2
p2
 
ω̇ 1
= 2 mω 2 x2 −
ω 2 2m
ω̇
= 2 (V (t) − K(t))
ω
Now
Z t+T
dI 0
I(t + T ) − I(t) = dt
t dt0
t+T
ω̇(t0 )
Z
= 2 (t0 )
(V (t0 ) − K(t0 )) dt0
t ω
Z t+T
ω̇(t)
≈ 2 V (t0 ) − K(t0 ) dt0
ω (t) t
≈0

where we justify the approximations as follows: ωω̇2 is approximately constant


over one period of oscillation. The resulting integrand vanishes for constant ω,
since we recall that kinetic and potential energy oscillate together with the same
period.
Let’s get a geometric picture for this result too. Consider the motion of our
system in phase space (x, p). Conservation of energy (for ω constant) gives

p2 1
+ mω 2 x2 = E,
2m 2
so the oscillator traces out an ellipse in the clockwise direction. This ellipse has
semiaxes

r
2E
a= 2
b = 2mE

2πE
and thus area πab = ω = 2πI. This area is just the contour integral
I
p dx (23.7)

along the phase space trajectory, so it appears that this expression is the “true”
adiabatic invariant: true for any oscillatory motion.

23.2 Quantum mechanical motivation


With our intuition, we look at the quantum harmonic oscillator. We compute
E 1
ω = h̄(n + 2 ), suggesting that quantum number n will be an adiabatic invariant:

69
Tony Zhang 23 April 6, 2017

that energy eigenstates will want to stay in an energy eigenstate under adiabatic
change.
We can motivate some more: for any one-dimensional Hamiltonian with turn-
ing points x1 and x2 for energy E, the WKB approximation gives quantization
condition I  
1 1
p(x) dx = 2π n + ,
h̄ 2
suggesting that quantum number n is an adiabatic invariant in the general case,
at least semiclassically.
Even more intuition. Recall that first-order time-dependent perturbation
theory gives Z 2
t 0
iωf i t0 δH f i (t )
f
0
Pr (t) = e dt . (23.8)

f ←i 0 ih̄
For a constant perturbation,
2
1 f 2 t iωf i t0 0
Z
Pr (t) = |δH |
fi e dt
f ←i h̄2 0

1 |eiωf i t |2
= 2 |δHf i |2
h̄ ωf2 i

so we see that a larger energy difference quadratically suppresses the transition


probability, which suggests that transitions are disfavored, especially as δH
shrinks.

23.3 Quantum adiabatic evolution


Suppose we have an instantaneous energy eigenstate |ψ(t)i:

H(t) |ψ(t)i = E(t) |ψ(t)i (23.9)

that is always normalized. Are we done? If we start in |ψ(0)i, will we always


stay in the stationary state?
Nope. The above equation is nice and useful, but it doesn’t tell us that our
state is stationary. In particular, if we start at |ψ(0)i, we have no guarantee we
will be in |ψ(t)i.
Indeed, the time-dependent Schrödinger equation gives

ih̄ |ψ(t)i = H(t) |ψ(t)i = E(t) |ψ(t)i (23.10)
∂t
which motivates the following ansatz as candidate wavefunction:
E(t0 ) dt0
i
Rt
|Ψ(t)i = c(t)e− h̄ 0 |ψ(t)i . (23.11)

Note that this ansatz isn’t completely general.


Plug into time-dependent Schrödinger equation:

|Ψ(t)i = H |Ψ(t)i
ih̄
∂t
i 0 i 0
R R
ċe− h̄ E dt |ψ(t)i + E(t) |ψ(t)i + c(t)e− h̄ E dt |ψ̇(t)i = E(t) |Ψ(t)i

70
Tony Zhang 23 April 6, 2017

so that
ċ(t) |ψ(t)i = −c(t) |ψ̇(t)i . (23.12)
Adjoining with the known instantaneous eigenstate gives

ċ(t) = −c(t) hψ(t)|ψ̇(t)i

so that
hψ|ψ̇i(t0 ) dt0
Rt
c(t) = c(0)e− 0 . (23.13)
Wait. This quantity looks like it’s real, which is bad. We want it pure imaginary,
so that |Ψi is properly normalized. We check:
Z
hψ|ψ̇i = ψ ∗ ψ̇ dx
Z  
d ∗
= (ψ ψ) − ψ̇ ∗ ψ) dx
dt
Z Z
d
= ψ ∗ ψ dx − ψ ψ̇ ∗ dx
dt
| {z }
1
Z ∗
=0− ψ ∗ ψ̇

= − hψ|ψ̇i

whence we have
E(t0 ) dt0 i hψ|ψ̇i dt0
i
Rt Rt
|Ψ(t)i = c(0)e− h̄ 0 e 0 |ψ(t)i . (23.14)

Wait. We’ve just proven that for arbitrary time-dependent Hamiltonians,


stationary states will still only evolve by phase. That can’t be right. What just
happened?
Look at how we proceded after Equation 23.12. Adjoining with hψ| lost us
information, since that adjoining is not an injective function. Everything that
followed was necessary, but not sufficient.
Indeed, if vectors |ψi and |ψ̇i are not parallel, which seems like something
that could totally happen, Equation 23.12 cannot be solved, meaning our ansatz
(Equation 23.11) is bad.
So is everything we did incorrect? Do we have to erase and start over? It
turns out the answer is no. Indeed, we will see that Equation 23.14 is true in
the adiabatic limit.
Before quitting, define some convenient notation:
i t
Z
θ(t) = − E(t0 ) dt0 (23.15)
h̄ 0
ν(t) = i hψ|ψ̇i (t) (23.16)
Z t
γ(t) = ν(t) dt0 (23.17)
0

so that
|Ψ(t)i = c(0)eiθ(t) eiγ(t) |ψ(t)i . (23.18)
We call the first phase (accumulated from the energy) the dynamical phase;
the second, the geometric phase or the Berry phase.

71
Tony Zhang 24 April 11, 2017

24 April 11, 2017


todotodotodotodo lost a section (apr 11)

72
Tony Zhang 25 April 12, 2017 (recitation) TODO TODO TODO TODO

25 April 12, 2017 (recitation) TODO TODO TODO


TODO
Recall in proving the adiabatic theorem that we neglected the term
X
c̃˙k = − hψk |ψ̇n i ei(θn −θk ) ei(γn −γk ) c̃n (t).
n6=k

We now prove we can do that.


Suppose the Hamiltonian varies in time from 0 ≤ t ≤ T as H(t). Define

H̃(t/T ) = H(t)

on the interval [0, 1]. To slow down the evolution of H, we would simply take
T → ∞. Assume H̃ has instantaneous eigenstates

H̃(s) |ψ̃n (s)i = Ẽn (s) |ψ̃n (s)i .

Now integrating our expression for c̃˙k , we get


XZ t ˙
c̃k (t) − c̃k (0) = hψ̃k |ψ̃n i ei(θn −θk ) ei(γn −γk ) c̃n (t0 ) dt0 (25.1)
n6=k 0

Let’s do some work on the phases.


Z t
γn (t) = i hψk |ψ̇k i dt0
0
Z t
∂ t0 dt0
=i hψ̃k | |ψ̃k i
0 ∂s T T
Z Tt

=i hψ̃k | |ψ̃k i ds
0 ∂s
≡ γ̃n (t/T )

Now
Z
1
θn (t) = − En (t0 ) dt0

1 t
Z
=− Ẽn (t0 /T ) dt0
h̄ 0
T t/T
Z
=− Ẽn (s0 ) ds0
h̄ 0
≡ T θ̃n (t/T )

This prefactor of T is our game-changer.


Therefore,
t
XZ ∂ t0 0 0 dt0
c̃k (t) − c̃k (0) = − hψ̃k | |ψ̃n i ( )ei(γ̃n −γ̃k )(t /T ) eiT (θ̃n −θ̃k )(t /T ) c̃n (t0 ) .
0 ∂s T T
n6=k

Now defining

Fkn (s) = hψ̃k | |ψ̃n i ei(γ̃n γ̃k )(s) ,
∂s
73
Tony Zhang 25 April 12, 2017 (recitation) TODO TODO TODO TODO

we rewrite as
XZ t
0
c̃k (t) − c̃k (0) = − Fkn (s0 )eiT (θ̃n −θ̃k )(s ) c̃n (s0 T )ds0
n6=k 0

Define
ĉ(s, T ) = c̃(sT )
so that
XZ t
0
ĉk (s, T ) − ĉk (0, T ) = − Fkn (s0 )eiT (θ̃n −θ̃k )(s ) ĉn (s0 , T ) ds0 .
n6=k 0

The key here will be to integrate by parts. In anticipation, we note that


d 1
θ̃n = − Ẽn (s)
ds h̄
whence
d iT (θ̃n −θ̃k )(s) iT
e = − (Ẽn − Ẽk )(s)eiT (θ̃n −θ̃k )(s) .
ds h̄
Now we invoke the assumption that there are no level crossings. Then rewrite
the integral

ih̄ X s Fkn (s0 )


Z
d 0
ĉk (s, T ) − ĉk (0, T ) = − ĉn (s0 , T ) 0 eiT (θ̃n −θ̃k )(s ) ds0
T 0 (Ẽn − Ẽk )(s0) ds
n6=k

and integrate by parts


 s

X ih̄  Fkn (s0 ) 0
ĉn (s0 , T )eiT (θ̃n −θ̃k )(s )

ĉk (s, T ) − ĉk (0, T ) = −
T (Ẽn − Ẽk )(s )
0
n6=k 0
s =0
ih̄ X s d Fkn (s0 )
Z  
0 iT (θ̃n −θ̃k )(s0 )
+ 0
ĉn (s , T ) e ds0
T ds (Ẽ − Ẽ )(s0)
n6=k 0 n k

Things are looking real good. We have lots of T −1 -prefactors. All the tilde’d
quantities depend on s ∈ [0, 1], so don’t have T -dependence.
d
Further, ĉk is bounded. All we want now is to show that ds ĉk (s, T ) is
bounded. Write:
d X
ĉk (s, T ) = − Fkn (s0 )eiT (θ̃n −θ̃k ) (s0 )ĉn (s0 , T )
ds
n6=k

But each quantity here is bounded, so ∆ĉk ∼ O(T −1 ). In fact, if H changes


smoothly (i.e. C ∞ ), Prk←n (T ) ≤ O(T −p ) for any exponent p.
Morally, this class of results holds for the same reason that the Fourier
transform for a C k function decays as O(T −k ).
TODO TODO TODO TODO

74
Tony Zhang 26 April 14, 2017

26 April 14, 2017


26.1 Landau-Zener transition probabilities
We continue our discussion of Landau-Zener transitions. Recall the setup: we
start with a spin- 12 system in a time-dependent magnetic field in the z direction,
such that 1 
2 αt 0
H= (26.1)
0 − 21 αt
with α > 0. Denote spin-up and spin-down states by |1i and |2i; these are
instantaneous eigenstates of our system. We saw that they can be upgraded to
solutions of the Schrödinger equation through the addition of a phase factor.
But in the presence of a slight perturbation to the Hamiltonian
1 
αt H12
H= 2∗ (26.2)
H12 − 21 αt

with time-independent constant H12 , the energy level crossing is avoided. The
new energies are given by
r
1
E± (t) = ± |H12 |2 + α2 t2 (26.3)
4
with an minimum energy gap of 2|H12 | at t = 0. We want to find the probability
of switching branches (the probability of a non-adiabatic transition, also
known as diabatic passage).
Assuming without loss of generality that H12 is real, the new instantaneous
eigenstates are precisely those pointed along the unit vector

(H12 , 0, 21 αt)
~n = q . (26.4)
2 + 1 α2 t2
H12 4

This vector flips as we move from t = −∞ to t = ∞. So an adiabatic change in


H will flip spin-up to spin-down and vice versa.

26.1.1 The adiabatic timescale

But recall that when |H12 | vanishes, we found that states do not flip at all, no
matter how slow you go. So we see that as the energy gap 2|H12 | shrinks, the
adiabatic timescale increases.
Let’s make that more precise. Holding α fixed, we see that the probability
of the non-adiabatic transition from the top branch to the bottom branch goes
to 1 as |H12 | → 0. Conversely, this probability is suppressed for large |H12 |.
In fact, define a timescale t∗ by using the “bounding box” of the hyperbola
(see Figure ??). We have
1
|H12 | = αt∗
2
or
|H12 |
t∗ = 2 ≡ 2τd . (26.5)
α
The quantity τd is a measure of the duration of the change in H.

75
Tony Zhang 26 April 14, 2017

But there’s another associated timescale with this system. At t = 0, the


resulting Hamiltonian is purely off-diagonal:
 
0 H12
H(t = 0) = ∗ (26.6)
H12 0

This Hamiltonian has Rabi frequency


|H12 |
ω12 = ; (26.7)

this is the frequency of flopping behavior of a system with time-independent
Hamiltonian H(t = 0). Morally, we see this frequency is a measure of the energy
gap 2|H12 | of the system.
Therefore, we declare that the change in H is adiabatic if

ω12 τd  1. (26.8)

Remark 26.1 (Alternative derivation of condition). We can also do the follow-


ing. Let T be the quantum time for the system: ω2π12
. We require that at all
times t,
dH(t)
T  H(t). (26.9)
dt
This condition is directly in analogy with the WKB condition that the de Broglie
wavelength changes little over a de Broglie wavelength:
 
d h̄
dx p  1.


Now rather unrigorously, write T = H and divide by H, so that

h̄ dH
H 2 dt  1.

As a result,  
d h̄
dt H(t)  1,

or  
d h̄
dt E(t)  1. (26.10)

26.1.2 Transition probabilities

At this point, we want to be more quantitative. What is the probability of


transitioning, as a function of α and |H12 |?
Writing  
E1 (t) H12
H(t) = ∗ ,
H12 E2 (t)
we make the ansatz
E1 (t0 ) dt0 E2 (t0 ) dt0
i
Rt i
Rt
Ψ = A(t)e− h̄ 0 |1i + B(t)e− h̄ 0 |2i . (26.11)

Suppose we start in state |2i in the infinite past, so A(t = −∞) = 0 and
B(t = −∞) = 1. We want to find B(t = ∞).

76
Tony Zhang 26 April 14, 2017

Using either special functions or complex analysis, one can find (as Zener
did) the transision probability: the probability of staying in |2i:

|B(t = ∞)|2 = e−2πω12 τd (26.12)

so we see our adiabatic-ness parameter ω12 τd involved in an exponential sup-


pression of the transition probability.

26.2 The Berry phase


Recall the adiabatic theorem, which tells us that we stay in instantaneous
eigenstates
H(t) |ψn (t)i = En (t) |ψn (t)i (26.13)
under adiabatic change, and that

|Ψ(t)i = eiθn (t) eiγn (t) |ψn (t)i (26.14)

with dynamical and geometric phases

1 t
Z
θn (t) = − En (t0 ) dt0 (26.15)
h̄ 0
Z t
γn (t) = νn (t0 ) dt0 (26.16)
0

where we have
νn (t) = i hψn (t)|ψ̇n (t)i . (26.17)

Berry’s observation was simple, but not realized for a long time. Suppose
now that H depends on N possibly time-dependent coordinates
~
R(t) = {R1 (t), . . . , RN (t)}

which parameterize a sort of configuration space. Again, suppose we have


instantaneous eigenstates
~ |ψn (R)i
H(R) ~ = En (R)
~ |ψn (R)i
~ . (26.18)

Now write the following:


 
d
νn (t) = i ψn (R(t)) ψn (R(t))
dt
n  
X d dRi
= i ψn (R(t)) ψn (R(t))
i=1
dRi dt
~
dR
= i hψn (R(t))| ∇~~ |ψn (R(t))i ·
R

|{z} 
dt
∂ ∂
∂R1 ,..., ∂R n

so that the geometric phase is


Z t ~
γn (t) = ~ 0 ))|∇
i hψn (R(t ~ 0 ))i · dR dt0
~ ~ |ψn (R(t
R
0 dt0

77
Tony Zhang 26 April 14, 2017

But this integral is simply a line integral in configuration space:


Z ~f
R
γn (t) = ~
i hψn (R)|∇ ~ ~
~ |ψn (R)i · dR (26.19)
R
~i
R

which no longer depends on time explicitly, only implicitly through the path
traced in configuration. In particular, if we evolve the Hamiltonian in the same
way through configuration space, just at different rates, the accumulation of the
geometric phase is the same.
Note that γn (R~i → R ~ f ), which in this form we call Berry’s phase, is
path-dependent. But there is no more time dependence. The N -component
integrand
~ n (R)
A ~ = i hψn (R)|∇
~ ~
r |ψn (R)i
~ (26.20)
is called Berry’s connection. We call it a connection rather than a vector
because it doesn’t quite behave like a vector; it’s like the electrodynamic vector
potential, which has a certain amount of gauge freedom.
Indeed, suppose we define new instantaneous eigenstates
~
~ = e−iβ(R) ~
|ψ̃n (R)i |ψn (R)i (26.21)

~ Uh oh. Now we’re going to get a new Berry connection:


for real phase angle β(R).

~ = i hψ̃n (R)|∇R |ψ̃m (R)i


Ãn (R)
= i hψn (R)|eiβ(R) ∇R e−iβ(R) |ψn (R)i
=A ~ n (R)
~ + ∇R β(R)

and therefore a new Berry phase:


Z Rf
γ̃n = ~ = γn + β(Rf ) − β(Ri ).
à · dR (26.22)
Ri

So our choice of guage will change the Berry phase. But if we trace a closed
path in configuration space, the phase becomes gauge-invariant.

78
Tony Zhang 27 April 20, 2017

27 April 20, 2017


27.1 Review
Recall that the adiabatic theorem shows that instantaneous energy eigenstates
can be used to form solutions to the time-dependent Schrödinger equation upon
iEt
multiplication by a dynamical phase (a generalization of e− h̄ ) and a geometric
phase Z t
γ= i hψn (t0 )|ψ̇n (t)i dt0 . (27.1)
0

It often happens that the time dependence of a Hamiltonian can be imagined


~
as its trajectory through some configuration space, parameterized by vector R.
~
We then showed that Berry’s phase becomes a line integral in R space, with no
remaining explicit time dependence:
Z Rf
γn (Ri → Rf ) = ~
i hψn (R)|∇ ~
~ ~
r |ψn (R)i ·dR. (27.2)
Ri | {z }
~ n (R)
A ~

where we defined a Berry connection A ~ n (R).


~ We’ll recall that n labels the
instantaneous energy eigenstates.
~
Now given |ψn (R)i, we’re totally free to multiply by some phase pointwise
~ to get new instantaneous eigenstates
at each R
~
~ = e−iβ(R) ~ .
|ψ̃n (R)i |ψn (R)i (27.3)

We showed that under such a “gauge transformation”,


~ f ) − β(R
γ̃n (Ri → Rf ) = γn (Ri → Rf ) + β(R ~ i) (27.4)

so that Berry’s phase over a closed path is gauge-invariant.

27.2 More on Berry’s phase


Some comments about Berry’s phase:

1. If the instantaneous eigenwavefunctions ψn (x, t) are real, Berry’s connec-


tion (and thus Berry’s phase) vanishes:

νn (t) = i hψn (t)|ψ̇n (t)i


Z
d
= i ψn∗ (t) ψn (t) dx
dt
Z
d
= i ψn (t) ψn (t) dx
dt
Z
i d 2
= ψn (x, t) dx
2 dt
| {z }
1
=0

2. If the configuration space is one-dimensional, Berry’s phase vanishes for


loops! Indeed, taking the line integral of the Berry connection (or any
quantity) along a closed loop gives zero.

79
Tony Zhang 27 April 20, 2017

3. In a three-dimensional configuration space, suppose we traverse loop Γ


forming the boundary of 2D surface S. Then
I ZZ
~ ~ ~
A(R) · dR = ~ · d~a.
(∇R~ × A) (27.5)
Γ

~
The curl on the right hand side is sometimes called Berry’s curvature B.

So we see that accumulating a Berry’s phase isn’t a trivial task.

Example 27.1
Consider an electron in a magnetic field
~ = B0~n(t).
B

The Hamiltonian is
H = −~ ~ = µB B0~n · ~σ
µ·B
where we use the Bohr magneton
eh̄
µB = .
2me c
~ is ~n on the unit sphere. Our
Here, the configuration space vector R
instantaneous eigenstates are

~n · ~σ |~ni = |~ni
~n · ~σ |−~ni = − |−~ni

Now suppose ~n changes adiabatically, traversing a closed loop. By adiabatic,


we mean that the timescale T of change is long:

T  .
µB B0
What geometric phase accumulation do we get? Perhaps we’ll do it in
recitation; the answer for the spin-up eigenstate is:
I
1
γ = i h~n|∇~n |~ni · d~n = − Ω (27.6)
2
where Ω is the solid angle subtended by the loop Γ. The answer for the
spin-down state is + 21 Ω.

27.3 Molecules and the Born-Oppenheimer approximation


As we already know, molecules are much more difficult to think about than
atoms. Why? We lose spherical symmetry, which basically rules out any analytic
solution. (Also, we now need to consider the positions of the nuclei as moving.)
So we go into an approximate picture. We have these nuclei with positive
charges. On their own, they would just repel each other and fly away. Fortunately,
we also have electrons; the electron clouds can sit between nuclei in a way that
stabilizes the system.

80
Tony Zhang 27 April 20, 2017

Still, it’s not super obvious, for example, whether two protons and an electron
can form a bound state. The answer to this question is actually yes: we can
prove the existence of a bound state, the H2 + ion.
Joke. “Mathematical physicists enjoy doing that kind of thing.”– (talking
about the rigorous proof that no bound state exists with three protons and one
electron)

The idea of the Born-Oppenheimer approximation is to consider the


nuclei in classical equilibrium due to repulsion and attraction induced by the
electrons. This approximation is justified because nuclei are much heavier
than electrons, and are thus almost fixed relative to the electrons. Indeed, the
electron-to-nucleus mass ratio is at most something like
me
∼ 10−4 ;
M
sometimes even 10−5 .
We now claim that slow nuclear vibrations adiabatically deform the electronic
states. To see, suppose the molecule has size approximately a. Then just by
dimensional analysis, we expect typical electronic momentum

Pe =
a
and energy
h̄2
Ee = .
ma2
Then assuming simple harmonic oscillations, the nuclear potential energy looks
something like
1
M ω 2 R2
2
where R is the amplitude of nuclear oscillation. We’d like to find ω, because
the vibrational energies will go like h̄ω (from the quantum harmonic oscillator).
If the oscillation amplitudes become comparable to a, then the vibrational
potential energy should be around the electronic confinement energy:

1 h̄2
M ω 2 a2 ∼ .
2 ma2
Note that we don’t expect nuclear oscillations to be so big. (But if they were,
we’d expect them to affect electron energy levels significantly.) (Yes, this is not
particularly rigorous, but meh.)
As a result,
1 h̄2 m h̄2
ω2 ∼ =
M ma4 M m2 a4
so
m h̄2
r r
m
Evib ∼ h̄ω = = Eelec .
M ma2 M

There’s yet another contribution to molecular energies, well-known in chem-


istry: molecular rotation. It turns out that the rotational energy is even smaller:

L2 h̄2 `(` + 1) m h̄2 m


Erot = ∼ ∼ = Eelec .
2I M a2 M ma2 M
81
Tony Zhang 27 April 20, 2017

It makes sense that this energy is smaller. Indeed, nuclear vibration slightly
deforms the electron clouds; this elasticity contributes.
The energy scales are then roughly
r
m m
Eelec : Evib : Erot = 1 : : . (27.7)
M M
So in a typical molecule, these energy scales each differ by two orders of magni-
tude.
We set up a molecular Hamiltonian now. We write
N n
X p2α ~ +
X p2i ~ ~r) + Vee (~r)
H= + Vnn (R) + Ven (R, (27.8)
α=1
2M α i=1
2m
| {z }
~
He (R)

where the terms are respectively nuclear kinetic, nucleus-nucleus interactions,


electron kinetic, electron-nucleus interactions, and electron-electron interactions.
Rα labels the positions of the N nuclei of masses Mα and momenta pα , while ri
gives the positions of the n electrons of mass m and momenta pi .
Joke. If it takes this long to write down this Hamiltonian, imagine how hard it
is to solve!

To think about this problem adiabatically, imagine we fix all nuclear degrees
of freedom. With a fixed nuclear skeleton, we solve for the electron wavefunctions.
These will precisely be our instantaneous eigenstates, and our configuration
space will be the nuclear degrees of freedom. Under “slow” vibrations of the
nuclei, the adiabatic theorem tells us that the electrons will stay in instantaneous
eigenstates.
In particular, we solve the Schrödinger equation
!
h̄2 X 2 ~ ~r) + Vee (~r) φ(i) (~r) = Ee(i) (R)φ
~ (i) (~r)
− ∇ri + Ven (R, R R (27.9)
2m i
This equation is still hard and intractible, but it’s much better than using the
full Hamiltonian from above. We can find candidate solutions by making guesses
(eh, this electron probably has a Gaussian-like distribution, parameterized by
some mean and covariance, etc.). If we want to estimate the ground state energy,
we can apply the variational method on our candidate wavefunction families,
finding the choice of parameters that minimize the energy expectation.
But suppose we went beyond even that and found analytic solutions, labeled
by superindex i, which gives the energy levels. Then we know the energies
(i) ~ (i)
Ee (R) and the wavefunctions φR (~r). To get the (approximate) solution for
the entire molecule, we would then write
~ (i) (~r).
X
~ ~r) =
ψ(R, η (i) (R)φ (27.10)
~
R
i

At this point we could plug into the full time-independent Schrödinger


equation for the molecule and solve for the coefficient functions η. But that’s
still hard, so we suppose that η has no i dependence! This is equivalent to
invoking the adaibatic approximation: we suppose that as nuclear positions R ~
changes, the relative amplitudes on the different electronic eigenstates remain
fixed.
We’ll talk about this more next time.

82
Tony Zhang 28 April 24, 2017 (recitation)

28 April 24, 2017 (recitation)


28.1 More about the Berry phase
We will talk more about the Berry phase. Recall that under adiabatic evolution
of the Hamiltonian, if we start in an energy eigenstate |ψn i, we remain in the
instantaneous eigenstates |ψn (t)i.
In adiabatic evolution, the state will accumulate a phase change:

eiγn (t) eiθn (t) |ψn (t)i (28.1)

where we have dynamical and geometric (or Berry) phases


Z t Z t
1
θn (t) = − E(t0 ) dt0 γn (t) = i hψn , ψ̇n i (28.2)
h̄ 0 0

If we suppose H traces a path in configuration space R(t), we can get Berry


phase I
~
γn = i hψn (R)|∇ ~ ~
R |ψn (R)i ·dR. (28.3)
| {z }
~
A(R)

where A ~ n is the Berry connection. Why did we restrict the integral to closed
paths? There is gauge freedom here: if we use instantaneous eigenstates
eiβ(R) |ψn (R)i, A~ n changes by ∇R β(R). If we trace a closed path in configuration
space, however, γn becomes gauge-independent.

28.1.1 Example: spin- 12 in B-field

Suppose we have an electron in a magnetic field B~n(t) where ~n is a slowly-varying


unit vector on the unit sphere. Suppose further that ~n eventually traces a closed
loop.
The instantaneous spin-up state is, expressing ~n with polar and azimuthal
angles θ and φ,
cos θ2
 
|+i = (28.4)
eiφ sin θ2
where the time-dependence of the angles is understood. We suppose we start in
|+(0)i.
Then because
1 1
∇R = ∂r r̂ + ∂θ θ̂ + ∂φ φ̂
r r sin θ
in spherical coordinates,

− sin θ2
     
0 1 1 0
∇R |+(t)i = r̂ + θ̂ + φ̂
0 2r eiφ cos θ2 r sin θ eiφ sin θ2

Note that the components of the gradient are spinors. Now adjoining h+(t)|
gives us the Berry connection
2 θ
~ + (R) = − 1 sin 2 φ̂ = − 1 tan θ φ̂.
A
r sin θ 2r 2

83
Tony Zhang 28 April 24, 2017 (recitation)

To integrate the connection over a closed loop in R3 , we integrate its curl


(the Berry curvature) over a surface with the loop as boundary. Find

~ + (R) = ∇ × A
~+ = 1 1 1
D ∂θ (Aφ sin θ)r̂ − ∂r (rAφ )θ̂ = − 2 r̂.
r sin θ r 2r
The Berry phase is then
ZZ Z
γ+ = D ~ = −1
~ + (R) · dΣ 1 1
r̂ · r2 dΩ r̂ = − Ω.
2 r2 2

Let’s explore the gauge freedom of the Berry connection. Consider an


alternate set of instantaneous spin-up states:
 −iφ
cos θ2

0 −iφ e
|+ (t)i = e |+(t)i = .
sin θ2

Redoing our fun times from before, we can compute gradient

−e−iφ sin θ2 −ie−iφ cos θ2


   
0 1 1
∇ |+ i = θ̂ + φ̂
2r cos θ2 r sin θ 0

and Berry connection


~ 0+ = 1 ~+.
A φ̂ + A
r sin θ
The correction here isn’t surprising. We added a phase −β(R) = −φ, so we
1
should get an extra term ∇R β = r sin θ φ̂, as we did.
Now taking a curl,
 
~+0 ~+ +∇× 1 ~+
D =D φ̂ =D
r sin θ

since the latter is the gradient of a function. We therefore get the same curl for
both.

84
Tony Zhang 29 April 25, 2017

29 April 25, 2017


29.1 Born-Oppenheimer: the nuclear wavefunction
We have

H = ĤN + Ĥe
X p2
ĤN = α ~
+ VN N (R)
α
2M α
X p2
Ĥe = i ~ ~r) + Vee (~r)
+ VeN (R,
i
2m

and time-independent Schrödinger equation

~ (i) (~r) = He φ(i) (~r)


Ee(i) (R)φ R R
!
h̄2 X 2 ~ ~r) + Vee (~r)
= − ∇ri + VeN (R,
2m i

where subscript α labels the N nuclei and subscript i labels the n electrons.
The V are the interactions between pairs of particles; we group the interaction
terms based on the types of the particles (i.e. electron or nucleus).
(i)
If we can solve the electronic wavefunctions φR (~r), the entire molecular
wavefunction is then
~ (i) (~r)
X
~ ~r) =
ψ(R, η (i) (R)φ ~
R
i

That is, we have an R ~ dependent superposition of φ(i) (~r). At this point, we


~
R
could solve for the η by plugging our (fully general) ansatz into the molecular
time-independent Schrödinger equation.
But that’s hard, so invoking the adiabatic approximation, we assume η is
independent of superindex i and write
~ ~r) = η(R)φ
ψ(R, ~ ~ (~r) (29.1)
R

At this point, we’ve made the Born-Oppenheimer approximation. The strategy


~
now is to ask: now that we know φR~ , how do we find η(R)?
This question is nontrivial, since we no longer have an exact solution of the
Schrödinger equation. Indeed, the nuclear wavefunctions η(R) ~ are not governed
just by ĤN : that would be purely repulsive. We’ll need to find a sort of “effective
potential”. Indeed, we expect some sort of dependence of the equations of motion
~
of η on electron energies Ee (R).

Remark 29.1. What we’re about to do is a baby version of the renormalization


group, a technique that allows us to work with a system at different energy
scales.

We use the variational method to approximate the ground state. We want


to choose η to minimize
~ ~r)|H|ψ(R,
hψ(R, ~ ~r)i (29.2)

85
Tony Zhang 29 April 25, 2017

~ ~r) = η(R)φ
where ψ(R, ~ ~ (~r). We claim that we can actually write this expectation
R
as
~
hη(R)|H ~
eff |η(R)i

for some effective Hamiltonian Heff .


Indeed, the Hamiltonian expectation is really
Z Z
~ ~r)|H|ψ(R,
hψ(R, ~ ~r)i = ~ ~ (~r)H(η(R)φ
η ∗ (R)φ ~ ~ (~r)) dR
~ d~r. (29.3)
R R

We shall be able to move some terms across the H and do the ~r integral.
Write hHi = hHN ifull + hHe ifull , where the subscripts denote expectations
~ ~r) Hilbert space, so that
on the (R,
Z Z
hHe ifull = d~r dR ~ η ∗ (R)φ
~ ∗R (r)He (η(R)φR (r))

But He commutes with η. To see, note that the kinetic term is a differential
operator in ~r only, so that commutes across η. Further, the potential terms are
multiplicative, so we can harmlessly move them. Therefore,
~ e φR (~r) = η(R)E
He (η(R)φR (r)) = η(R)H ~ e (R)φ
~ R (~r)

and
Z Z
hHe ifull = d~r ~ η ∗ (R)φ
dR ~ ∗R (r)η(R)E
~ e (R)φ
~ R (~r)
Z
= ~ η ∗ (R)
dR ~ Ee (R)
~ η(R)
~
| {z }
∈Heff

where we’ve performed the ~r integral.


~ with something else. Yay, we’ve
So we see that Heff will be the sum of Ee (R)
just done the easy term. Now let’s move on to HN . The VN N component is
multiplicative, so it commutes across η as well in the expectation integral, and
Z
hVN N (R)ifull = η ∗ (R) VN N (R) η(R) dR. (29.4)
| {z }
∈Heff

Finally, we consider the nuclear kinetic term, the really nontrivial portion.
Indeed, it contains derivatives of R, so we can’t just commute it across η. We
write
 2  Z
pα 1 ~ ∗ (~r)pα pα (η(R)φR (~r)) dr dR
= η ∗ (R)φ R
2Mα full 2Mα

where we recall pα = −ih̄∇R . (As we mentioned before, we can’t move either


φR factor across the pα pα operator.)
We won’t go into super full detail, but we will do enough so that you can fill
in the details. Let us continue with the product rule:
 2  Z
pα 1 ~ ∗R (~r)pα [(pα η(R))φR (r) + η(R)(pα φR (~r))] dr dR
= η ∗ (R)φ
2Mα full 2Mα

86
Tony Zhang 29 April 25, 2017

Now we have
φ∗R pα = pα φ∗R − (pα φ∗R )
where every term should be thought of as an operator. Then we get four terms
in our expectation
 2  Z
pα 1 ~
= dr dR η ∗ (R)
2Mα full 2Mα
{pα [φ∗R (pα η(R))φR (r) + φ∗R η(R)(pα φR (~r))]
−(pα φ∗R (r)) [(pα η(R))φR + η(R)(pα φR (r))]}

We would like to integrate away the r dependence. Shuffle terms around:


 2  Z
pα 1 ~
= dR η ∗ (R)
2Mα full 2Mα
  Z Z 
∗ ∗
pα (pα η(R)) φR (r)φR (r) dr + η(R) φR (pα φR (~r)) dr
Z Z 
∗ ∗
−(pα η(R)) (pα φR (r))φR (r) dr − η(R) (pα φR (r))(pα φR (r)) dr

where, as good physicists, we make all the simplifications we know how to make
and give the things we don’t know
R how to deal with a name. We see a 1: good!
But we see a Berry-like thing: φ∗ (pα φ) dr. Indeed, it’s a Berry connection
Z Z
~ = ih̄ φ∗R (r)∇R φR (r) dr = − φ∗R (r)pα φR (r) dr
Aα (R) (29.5)
α

~ configuration space.
over the R
Continue:
p2α
  Z
1 ~
= dR η ∗ (R)
2Mα full 2Mα
 2
pα η(R) − pα (η(R)Aα (R))
Z 
2
−Aα (R)pα η(R) + h̄2 η(R) |∇R φ∗R (r)| dr

Joke. Well I ended up doing the whole calculation.

Now we can simplify to something downright nice:

p2α (pα − Aα )2 A2α h̄2


  Z  Z 
∗ ~ 2 ~
= dR η (R) − + |∇R φR (r)| dr η(R)
2Mα full 2Mα 2Mα 2Mα
(29.6)
where it is understood that (p − A)2 = p2 − pA − Ap + A2 .
Finally, we get the effective Hamiltonian:
" #
Z X (pα − Aα )2
hHi = dR η ∗ (R) + U (R) η(R) (29.7)
α
2Mα
| {z }
Heff

where
X h̄2 Z 2
X A2
α
U (R) = VN N (R) + Ee (R) + |∇R φR (r)| dr − . (29.8)
α
2M α α
2M α

87
Tony Zhang 29 April 25, 2017

For comparison, the naive effective Hamiltonian is:


X p2
α
+ VN N (R) + Ee (R). (29.9)
α
2M α

While our Heff is conceptually nice, it’s still real hard to solve. In practice,
people often start with the naive Heff , bringing in the full effective Hamiltonian
if they need more accurate answers.

29.2 Example: the H2 + ion


The H2 + ion consists of two protons separated by R with an electron at
distance ri from proton i. The Hamiltonian is
2
p2α h̄2 2
 
X
2 1 1
H= + VN N (R) − ∇ −e + . (29.10)
α=1
2M 2m r r1 r2
| {z }
He

How do we Born-Oppenheimer this problem? Let us consider the electronic


Hamiltonian He . How hard is it to find eigenstates? Still pretty hard, because
we don’t have spherical symmetry.
Therefore, we use the variational method. Consider trial wavefunction
ψtrial = A [ψ0 (r1 ) + ψ0 (r2 )] (29.11)
where
1
ψ0 = p e−r/a0 (29.12)
πa30
is the ground-state wavefunction of hydrogen. (We’re using the LCAO technique:
linear combinations of atomic orbitals.)
The nice thing about our ansatz is that it’s analytic and mathematically
friendly. Furthermore, when R is large, the true ground state wavefunction
probably does resemble ψ.
We’d like energies as a function of R. To do so, we normalize our ansatz:
1
|A|2 =
2(1 + I)
where "  2 #
−R R 1 R
I=e a 1+ + .
a 3 a
Then the expected electronic energy is

h̄2 ∇2
   
1 1
hHe i = ψtrial − − e2 + ψ trial = Ee (R) (29.13)
2m r1 r2

Now the total potential is just VN N (R) + Ee (R) which gives a minimum
energy of −1.76 eV at R = 1.3 Å. The true answer is E = −2.8 eV at R = 1.06 Å.
Why the poor agreement? Well, our ansatz must be seriously wrong in some
way; indeed, when R → 0, we just get the regular hydrogen ground state
wavefunction. But because the nuclear charge is 2e, the real ground state should
be the hydrogen ground state, contracted toward the origin by a factor of 2.

88
Tony Zhang

Part VI

Scattering

89
Tony Zhang 30 April 26, 2017 (recitation)

30 April 26, 2017 (recitation)


30.1 Scattering warm-up
Consider the three-dimensional free-particle Hamiltonian

p2
H=
2m
(so V = 0). The eigenstates are
~
ψk ∝ eik·~x

with energies
h̄2 k 2
E=
.
2m
But our system is rotationally invariant, so we could alternatively use spherical
coordinates, giving eigenstates of the form

ψn`m = Rn` (r)Y`m (Ω).

It will be very useful to be able to move between these two eigenbases. Thus,
we’d like to find expansion “coefficients” C such that
~
X
eik·~x = C~k,`,m (r)Y`m (Ω).
`,m

Without loss of generality, choose ~k = kẑ. Then we claim there’s a nice formula:

ikz ikr cos θ
√ X √
e =e = 4π i` 2` + 1j` (kr)Y`,0 (Ω). (30.1)
`=0

Here, we invoke the spherical Bessel functions


 `
` ` 1 ∂ sin ρ
j` (ρ) = (−1) ρ . (30.2)
ρ ∂ρ ρ

Let’s prove this result. Defining R(r) = u(r)


r for a spherical eigenstate, we
have the radial equation

h̄2
 2 
∂ u `(` + 1)
− 2 + u = Eu.
2m ∂r r2
h̄2 k2
Invoking E = 2m ,
∂ 2 u `(` + 1)
+− u = k 2 u.
∂r2 r2
This is the radial equation corresponding to the Helmholtz equation in spherical
coordinates, whose solutions u/r are superpositions of spherical Bessel func-
tions j` and n` , where n` is j` except with a cosine. Since n` blows up near 0,
we forbid its presence, so that

ψk`m (r) ∝ j` (kr).

90
Tony Zhang 30 April 26, 2017 (recitation)

(Apparently there’s a cute way to get all solutions from the ` = 0 one by acting
with raising opeator p+ = px + ipy .)
Now we can express a Cartesian eigenstate as a superposition of the spherical
eigenstates of the same eigenvalue:
X
eikr cos θ = C`,m j` (kr)Y`m (Ω).
`,m

Acting with Lz , we see that the left hand side has eigenvalue 0; therefore the
Y`,m terms vanish unless m = 0, and we write C`,m = C` δm0 and
X
eikr cos θ = C` j` (kr)Y`0 (Ω). (30.3)
`

Hold r fixed and take an inner product with another m = 0 spherical


harmonic so that Z

C` j` (kr) = Y`,0 (Ω)eikr cos θ dΩ. (30.4)

Now it is well-known that


 `
1 L−
Y`,0 = p Y`,`
(2`)! h̄

where we’ve used a ladder operator. Plugging in above,


Z  `
1 ∗ L+
C` j` (kr) = p Y`,` eikr cos θ dΩ. (30.5)
(2`)! h̄

Now in spherical coordinates,


 
iφ ∂ ∂
L+ = h̄e + i cot θ . (30.6)
∂θ ∂φ

We claim that
 n
L+
eikr cos θ = (−ikr)n sinn θeinφ eikr cos θ . (30.7)

This fact can be proven by induction. Then


 `
L+
eikr cos θ = (−ikr)` sin` θei`φ eikr cos θ


2` `! 4π
=p (ikr)` Y`,` (Ω)eikr cos θ .
(2` + 1)!

Plugging into Equation 30.5 and using orthonormality of spherical harmonics


gives s

C` j` (kr) = 2` `! (ikr)` eikr cos θ (30.8)
(2` + 1)! (2`)!

Write ρ = kr and we consider the ρ → 0 limit. It can be shown that

2` `!
j` (ρ) → ρ` . (30.9)
(2` + 1)!
91
Tony Zhang 30 April 26, 2017 (recitation)

to lowest order in ρ, so to order `,


s
2` `! 4π
C` ρ` = 2` `! (iρ)`
(2` + 1)! (2` + 1)! (2`)!

and p
C` = i` 4π(2` + 1). (30.10)
Our desired result follows from plugging these coefficients back into the expansion
formula:

√ X √
eikz = eikr cos θ = 4π i` 2` + 1j` (kr)Y`,0 (Ω). (30.11)
`=0

92
Tony Zhang 31 April 27, 2017

31 April 27, 2017


31.1 Scattering
Scattering is basically where particles hit other particles. It’s super important,
because we do scattering experiments all the time, where we aim a beam at a
target surrounded by detectors.
In general, particle collisions are like reactions:

p + p −−→ p + p + π0
p + p −−→ p + n + π+
e+ + e− −−→ μ+ + μ−

In scattering, the “reactants” and the “products” are identical. In elastic


scattering, none of the particles’ internal states change in the collision.
We will focus on

• Elastic scattering
• Spinless particles
• Non-relativistic scattering
• Interactions V (~r1 − ~r2 ) that we can turn into one-body problems by going
into the center-of-mass frame and using reduced masses.
• Energy eigenstates

As before, we want to solve the free particle Schrödinger equation

h̄2 2
 
− ∇ + V (~r) ψ(~r) = Eψ(~r).
2M

For a scattering setup, we assume the potential has finite range a: that is, it
vanishes faster than r−1 for r → ∞. We look at scattering states: the solutions
of positive energy
h̄2 k 2
E= .
2M
So we want to solve
h̄2
 
∇2 + k 2 + V (~r) ψ(~r) = 0.


2M

The resulting solutions actually have infinite degeneracy— just consider the case
where V = 0.
Now in one-dimensional scattering, we would always posit an incident solution
Aeikx with reflection Be−ikx and transmission Ceikx . We’d like to do something
similar. Unfortunately, in 3D we don’t have a “left” and “right” side of the
potential.
Still, consider incident wave eikz . We can’t really talk about reflection and
transmission any more, so we just consider the scattered wave as one thing.
Naively, we ask whether a spherically symmetric outgoing wave eikr can be
considered a candidate scattered wave?

93
Tony Zhang 31 April 27, 2017

Actually, no, because it doesn’t solve the free-particle radial equation. But
we can divide by r to get a solution:

eikr
(∇2 + k 2 ) =0
r
which works for r 6= 0. But eikr /r doesn’t product angular dependence, which
is definitely something we want. So write

eikr
ψs (~x) = f (θ, φ)
r
with direction-dependent scattering amplitude f . Including fk makes this
solution only asymptotically: when r  a. All together,

f (θ, φ)eikr
ψ(~r) = φ(~r) + ψs (~r) = eikz + .
r
We expect fk to be determined by the scattering potential.
Now we define the differential cross-section dσ as
particles scattered per unit time into solid angle dΩ
dσ = .
incident particle flux density
Intiutively, the cross-section gives an effective size of the scattering potential.
To compute cross-sections, let’s talk about probability currents. The denom-
inator is the incident probability current
h̄ h̄k
=(ψ ∗ ∇ψ) = ẑ
m m
where ψ = eikz . We can get this result also by intuitively observing that
probability current is just probability density multiplied by v:
p h̄k
|eikz |2 × = .
m m

To find the numerator, we need the number of particles in volume element


dr dΩ:
f (θ, φ)eikr 2 2

r dr dΩ = |f (Ω)|2 dr dΩ.
r

Now scattered particles spend dt = drv time in volume element dr dΩ. So the
number of particles per unit time going through the volume element is

|f (Ω)|2 dr dΩ h̄k
dr
= |f (Ω)|2 dΩ
v
m

But we recognize the latter as incident flux, so that

|f (Ω)|2 dΩ h̄k
dσ = h̄k
m
= |f (Ω)|2 dΩ.
m

As a result, we see that the scattering amplitude is a pretty good name for f .
Finally, we can define a total cross-section σ as the integral of all the dσ. Our
problem now is to look for scattering amplitude f (Ω).

94
Tony Zhang 31 April 27, 2017

31.2 Partial waves


Suppose we have a spherically symmetric potential V . Then we want spherically
symmetric energy eigenstates. The following was covered in recitation, so we’ll
briefly summarize.
For fixed angular momentum quantum number `, we can posit solution
ψ = ur` Y`m . From recitation, we know that ur` is a superposition of spherical
Bessel functions j` and n` , which asymptotically look like
 
1 1
j` (x) ∼ sin x − `π
x 2
 
1 1
n` (x) ∼ − cos x − `π
x 2
for x → ∞. j` is regular at r = 0, while n` is singular. So any regular
energy eigenstate with angular momentum m = 0 must be a superposition of
wavefunctions ur` Y`,0 for various values of `.
As we saw in recitation,

√ X √
eikz = eikr cos θ = 4π i` 2` + 1j` (kr)Y`,0 (θ). (31.1)
`=0

We can prove Equation 31.1 with the relations.


r Z 1
2` + 1 1
Y`,0 (θ) = P` (cos θ) j` (z) = ` eizu P` (u) du. (31.2)
4π 2i −1
What’s wavy about the superposition on the right hand side of Equation 31.1?
The waviness lies in the j` spherical Bessel functions, as we can see in the
expression for j` .
Anyway, for large r, we can plug the asymptotic form of j` in to Equation 31.1
to get
√ i(kr− 12 `π) −i(kr− 12 `π)
!
4π X √ 1 e e
eikz ' i` 2` + 1Y`,0 (θ) − (31.3)
k 2i r r
`

so we identify an outgoing wave eikr /r and an ingoing wave e−ikr /r.

31.3 Phase shifts


Briefly, we recall the theory of one-dimensional scattering. Suppose we have an
infinite potential wall for x < 0, and V = 0 elsewhere.
 Then we could decompose
energy eigenstates φ(x) = sin kx = 2i 1
eikx − e−ikx into outgoing and ingoing
components.
If V was instead a nontrivial potential of finite range a we could still write
energy eigenstates exactly, in the region x > a, as
1 ikx+2iδ
ψ(x) = (e − e−ikx )
2i
for some phase shift δ. Indeed, conservation of probability requires that the
outgoing wave differ from the ingoing wave by only a phase. In the one-
dimensional case, we defined
ψ(x) = ψs (x) + φ(x).

95
Tony Zhang 31 April 27, 2017

Now back in three-dimensions, we would like to write an actual ansatz with


asymptotic form
eikr
ψ(r) ' eikz + fk (Ω) (31.4)
r
for r  a.
The only ingoing probability current contribution comes from the eikz term.
Then in analogy with the one-dimensional case, we take the asymptotic form of
the spherical wave expansion of eikz (Equation 31.3) and add phase shifts δ` to
the outgoing components to get an ansatz for ψ(r):
√ 1 1
!
4π X ` √ 1 ei(kr− 2 `π) e2iδ` e−i(kr− 2 `π)
ψ(r) = i 2` + 1Y`,0 (θ) − (31.5)
k 2i r r
`
eikr
= eikz + fk (θ) (31.6)
r

If we expand eikz as spherical waves in the large r approximation, we get


an ingoing contribution that matches that of the ansatz. Now moving the eikz
outgoing component across the equal sign,
√ 1
eikr 4π X ` √ e2iδ` − 1 ei(kr− 2 `π)
fk (θ) = i 2` + 1Y`,0 (θ)
r k 2i r
`

4π X √ eikr
= 2` + 1Y`,0 (θ)eiδ` sin δ`
k r
`

It follows that phase shifts are related to the scattering amplitude by



4π X √
fk (θ) = 2` + 1Y`,0 (θ)eiδ` sin δ` . (31.7)
k
`

96
Tony Zhang 32 May 1, 2017 (recitation)

32 May 1, 2017 (recitation)


32.1 The optical theorem
Today we prove the optical theorem, which gives the scattering cross-section of
some potential with respect to incident scattering state |ψ~k i as

σ= =(fk (θ, φ)). (32.1)
k
We know that three-dimensional probability current is given as

~ = =(ψ ∗ ∇ψ). (32.2)
m
Assuming ingoing plane wave eikz , the incident probability current is then
h̄k
~in = ẑ. (32.3)
m
For θ 6= 0 (away from the direction of propagation of the incident wave), the
wavefunction is just ψ~k = ψs , the scattered wave, so we have

~out = =(ψs∗ ∇ψs )
m
∗  !
eikr eikr


= = fk (Ω) ikfk (Ω) r̂
m r r
h̄k
= |fk (Ω)|2 r̂ (32.4)
mr2
Now for θ ≈ 0, the wavefunction includes the original incident wave. Thus,
eikr
 
ikz
∇ψk = ik e ẑ + fk (Ω) r̂
r
where we truncate higher powers of r, using the fact that ∇ = r̂∂r + O(r−1 ) in
spherical coordinates.
We bash to get

~out = =(ψk∗ ∇ψk )
m
e−ikr eikr
  
h̄k
= < e−ikz + fk∗ (Ω) eikz ẑ + fk (Ω) r̂
m r r
2 ik(r−z)
e−ik(r−z)
 
h̄k h̄k |fk (Ω)| h̄k e ∗
= ẑ + 2
r̂ + < fk (Ω) r̂ + f k (Ω) ẑ
m
|{z} |m {zr } |m r r
{z }
~
unscat ~
scat ~
inter

We invoke conservation of probability flux on a radius r sphere:


Z Z
~in = ~out · dS
~in · dS ~out
Z Z
~ ~out
~unscat · dSin = (~unscat + ~scat + ~inter ) · dS

|fk (Ω)|2 2
Z Z
h̄k ~out
0= r dΩ + ~inter · dS
m r2
| {z }
σ

97
Tony Zhang 32 May 1, 2017 (recitation)

What is the flux contribution from ~inter ? Assuming large r, we have r̂ ≈ ẑ and
θ ≈ 0:
eik(r−z) e−ik(r−z)
Z Z  
~ h̄k ∗ ~out
~inter · dSout = < fk (Ω) r̂ + fk (Ω) ẑ · dS
m r r
eik(r−z) e−ik(r−z)
Z  
h̄k ∗
= < fk (θ = 0) + fk (0) dx dy
m r r

Because
r sin2 θ 1 1 x2 + y 2
r − z = r(1 − cos θ) = ≈ r sin2 θ =
1 + cos θ 2 2 r
we can continue
ik(x2 +y 2 )
Z Z !
~out 2h̄k e 2z
~inter · dS = < fk (0) dx dy
m z
 Z ∞ 
2h̄k 2π ikρ2 /2z
= < fk (0) ρe dρ
m z 0

where we’ve gone into cylindrical coordinates and integrated away the azimuthal
ρ2
angle. Now substitute u = 2z to write
Z  Z ∞ 
~out = 4πh̄k < fk (0)
~inter · dS eiku du
m 0

But the integral looks divergent. For a quick fix, we regulate it as:
Z ∞
1 i
ei(k+iε)u du = − =
0 i(k + iε) k + iε

so we say the integral is ki . (Actually, we should be more careful and use


wavepackets instead of unnormalizable plane waves to avoid these divergences.)
Then
Z
~out = − 4πh̄
~inter · dS =(fk (0))
m
so that

σ= =(fk (0)). (32.5)
k

98
Tony Zhang 33 May 2, 2017

33 May 2, 2017
33.1 More about phase shifts
Recall our progress last time. We wanted a nice asymptotic form for the scattered
wavefunction for large r (far form the scattering center) in the case of a central
potential. We wrote
eikr
ψ(~r) ' eikz + f (θ) (33.1)
r
for r  a, where a is the range of the spherically symmetric potential, which we
assume drops off faster than r−1 .
Now the free-particle Hamiltonian commutes with momentum and with angu-
lar momentum. But linear and angular momenta don’t commute! Diagonalizing
momentum with the Hamiltonian gives plane waves as eigenbasis; choosing
angular momentum gets us spherical waves.
We want to write ψ as a superposition of an ingoing spherical wave and an
outgoing one. The ingoing wave is fixed, since only the eikz term contributes to
ingoing probability current. That fixes the outgoing wave, by conservation of
probability, and we got a sum over ` of terms containing partial waves
`π `π
ei(kr− 2 +2δ` ) e−i(kr− 2 )
− . (33.2)
r r
In essence, we’re parameterizing our ignorance in two ways: either with f (which
is what we’re looking for) or with phase shifts δ` . It turns out that we can relate
the two by √ ∞
4π X √
f (θ) = 2` + 1Yl,0 (θ)eiδ` sin δ` . (33.3)
k
`=0

The differential cross section, of course, is still



= |f (θ)|2 , (33.4)
dΩ
so we can integrate to get
Z
σ = |f (θ)|2 dΩ
Z
= f ∗ (θ)f (θ) dΩ
4π X √ √
Z
= 2` + 1 2`0 + 1e−iδ` sin δ eiδ`0 sin δ 0 ∗
Y`,0 (Ω)Y`0 ,0 (Ω) dΩ
` `
k2 0
`,` | {z }
δ`,`0

and
4π X
σ= (2` + 1) sin2 δ` . (33.5)
k2
`

In recitation yesterday we presented a slick proof of the optical theorem,


which gives σ in terms of f (θ = 0). We’ll prove it again today with a remarkably
simple brute-force method.

99
Tony Zhang 33 May 2, 2017

First, we present some useful facts about spherical harmonics:


r
2` + 1
Y`,0 (θ) = P` (cos θ) (33.6)

r
2` + 1
Y`,0 (θ = 0) = (33.7)

where P` is a Legendre polynomial. But
√ ∞ r
4π X √ 2` + 1 iδ`
f (θ = 0) = 2` + 1 e sin δ`
k 4π
`=0

1X
= (2` + 1)eiδ` sin δ`
k
`=0

1X
=(f (θ = 0)) = (2` + 1) sin2 δ`
k
`=0
1 k2 σ
=
k 4π
whence the optical theorem follows:

σ= =(f (0)). (33.8)
k

Now we restrict ourselves to consider a fixed ` (fixed angular momentum


multiplet), so that we can work with the radial equation

d2 u `(` + 1)
− u = −k 2 u. (33.9)
dr2 r2
If the potential vanishes outside some finite range a, the solution u/r for r > a
is
ψ(x)|` = (A` j` (kr) + B` n` (kr))Y`,0 (θ). (33.10)
Alternatively, if the potential decays sufficiently rapidly, ψ is a solution far away.
Basically, this ψ gives the form of a solution sufficiently far from the origin.
The B` term is a sign of scattering! Indeed, when there is no potential
(V = 0), this solution should hold all the way to the origin. But the spherical
Bessel functions n` are singular at r = 0. Therefore, their presence indicates
that there’s a potential making the solution near the origin something else that
eventually becomes ψ far out.
For kr large, we take the asymptotic expansions of the Bessel functions:
" #
sin(kr − `π
2 ) cos(kr − `π
2 )
ψ(x)|` ' A` − B` Y`,0 (θ). (33.11)
kr kr

Now we claim that


B`
tan δ` = − . (33.12)
A`
We write our ψ asymptotically as a superposition of an ingoing and an outgoing

100
Tony Zhang 33 May 2, 2017

wave. Write
    
1 `π `π
ψ(x)|` ∝ sin kr − + tan δ` cos kr − Y`,0 (θ)
kr 2 2
    
1 `π `π
= cos δ` sin kr − + sin δ` cos kr − Y`,0 (θ)
kr cos δ` 2 2
 
1 `π
= sin kr − + δ` Y`,0 (θ) (33.13)
kr 2

so that
1  i(kr− `π +δ` ) `π

ψ(x)|` ∝ e 2 − e−i(kr− 2 +δ` ) Y`,0
2ikr
e−iδ`  i(kr− `π +2δ` ) `π

= e 2 − e−i(kr− 2 )
2ikr

Our discussion shows that one can get the phase shift by any of three ways.

1. Use Equation 33.12 with the Bessel function solution outside the potential
range.

2. Read δ` off of asymptotic form Equation 33.13.


3. Recognize δ` by decomposing a solution asymptotically into ingoing and
outgoing waves (Equation 33.2).
Remark 33.1 (Reading suggestions). For scattering, Griffiths is OK. Merzbacher
is advanced but nice as usual. Cohen-Tannoudji also nice. Shankar is probably
OK too, if Professor Zwiebach recalls correctly.

33.2 Example: hard sphere scattering


Consider the hard sphere potential
(
∞ r≤a
V (r) = . (33.14)
0 r>a

Because V vanishes for r > a, Equation 33.10 gives the exact radial solutions
u
= R` (r) = A` j` (kr) + B` n` (kr). (33.15)
r
A general solution is then a superposition
X
ψ(r, θ) = R` (r)P` (cos θ) (33.16)
`

obeying boundary condition


X
ψ(a, θ) = R` (a)P` (cos θ) = 0 (33.17)
`

where the P` are Legendre polynomials. But these are complete, so R` (a) = 0
for all `:
A` j` (ka) + B` n` (ka) = 0. (33.18)

101
Tony Zhang 33 May 2, 2017

Then
B` j` (ka)
tan δ` = − = (33.19)
A` n` (ka)
which technically qualifies as having solved the problem, in that we understand
the Bessel functions well and that found the phase shifts.
Let’s find the scattering cross-section. The trig identity

tan2 δ`
sin2 δ` = . (33.20)
1 + tan2 δ`
gives
j`2 (ka)
sin2 δ` = (33.21)
j` (ka) + n2` (ka)
2

and thus

4π X j`2
σ= (2` + 1) (ka). (33.22)
k2 j`2 + n2`
`=0

In the case of low-energy scattering ka  1 (i.e. with wavelengths much


larger than the sphere), leading order expansions of j` and n` for small arguments
tell us that  ` 4
4π X 1 2 `!
σ≈ 2 (ka)4`+2 . (33.23)
k 2k + 1 (2`)!
`

See Griffiths for a derivation.


If the sum is tractable, this expansion is nice. In particular, if the terms fall
off rapidly with `, we’re good. But that’s exactly the case when ka  1, because
then the powers of ka drop fast. In fact, it often suffices simply to consider the
dominant ` = 0 contribution:

σ= (ka)2 = 4πa2 . (33.24)
k2
Again, we emphasize that this result holds in the low-energy limit: either we
have a tiny sphere or a tiny energy. Note that σ is the surface area of the entire
sphere, four times larger than the cross-sectional area πa2 . It appears that in
low-energy scattering, the wave can “feel around” the potential. Anyhow, no
one has really come up with good intuition for this phenomenon.
One more calculation we’d like to do: find e2iδ` . It’s basically a trig bash.
We have the identity

eiδ − e−iδ e2iδ − 1


i tan δ = iδ −iδ
= 2iδ (33.25)
e +e e +1
which gives
1 + i tan δ
e2iδ = . (33.26)
1 − i tan δ
As a result,
2iδ j` − in`
e =− . (33.27)
j` + in` r=ka

This example is pretty illustrative of the sorts of things we do a typical


scattering problem.

102
Tony Zhang 33 May 2, 2017

33.3 General computation of the phase shift


Suppose we start with a radial solution R` (r) known for r < a. Assume that
V (r > a) = 0. We want to extend R` past a as superposition

A` j` (kr) + B` n` (kr).

We enforce boundary conditions:

R` (a) = A` j` (ka) + B` n` (ka) (33.28)


aR`0 (a) = (ka)(A` j`0 (ka) + B` n0` (ka)) (33.29)

The convenient way to solve this equation is to form the logarithmic derivative
aR`0 (a) A` je0 ll + B` n0`
 0
j` − tan δ` n0`

β` (a) ≡ = ka = ka (33.30)
R` (a) A` j` + B` n` j` − tan δ` n`
At this point, we’re done in principle; it suffices to solve for tan δ` .
But sometimes people want to find e2iδ` . Unenlightening algebra gives
ka(j`0 − in0` ) − β` (j` − in` )
e2iδ` = − . (33.31)
ka(j`0 + in0` ) − β` (j` + in` )
hs
Writing Equation 33.27 as e2iδ` , it turns out that
 
j` −in`
hs β ` − ka( j` −in` )
e2iδ` = e2iδ`  j 0 +in0
(33.32)
β` − ka( j`` −in`` )

33.4 Intuition for phase shifts


We’ll provide some intuition for phase shifts. Our arguments will be semiclassical
and rather handwavy.

33.4.1 Impact parameters

Recall that we showed that in low-energy scattering, the lowest ` contributions


are largest. How does angular momentum show up in scattering?
Suppose (classically) that we have central potential of finite range a and
an incoming particle with impact parameter b and momentum p, thus angular
momentum L = bp. But quantum mechanically, L ≈ h̄`. Finally, p = h̄k, so
`
b∼ . (33.33)
k
Classically, we expect no scattering if b > a, so we don’t expect scattering if
` > ka. Indeed, only ` < ka typically contribute significantly.
To make this intuition a little more rigorous, consider free particle partial
waves
ψ` = j` (kr)Y`,0 (θ). (33.34)
For fixed θ, the probability to find the particle at radius r is proportional to

(kr)2`+2
j` (kr)r2 , dr dΩ ∼ dr dΩ. (33.35)
(2` + 1)! !
103
Tony Zhang 33 May 2, 2017

where we take the asymptotic form near the origin. (Far away from the origin,
j`2 r2 becomes a sine wave, as we have seen.) This asymptotic form is really tiny
until kr ≈ 1, at which point it takes off. The idea is that higher order partial
waves only become nontrivial somewhat far from the origin.
Remark 33.2 (Found in official class notes). In fact, considering scattering
scate with some k, the turning point can be found by equating energy with the
effective potential:
h̄2 `(` + 1) h̄2 k 2
= (33.36)
2mr2 2m
which gives a turning point r ≈ k` . We expect the solution j` (kr) to be
exponentially small within this turning point (i.e. in the classically foribdden
zone).

33.4.2 Partial wave unitarity

Recall

4π X 2
X
σ= (2` + 1) sin δ ` ≡ σ` . (33.37)
k2
`=0 `

We can thus bound


4π 4π
σ` ≡ (2` + 1) sin2 δ` = 2 (2` + 1). (33.38)
k2 k
This bound gives the partial wave unitarity bound. Basically, it enforces
conservation of probability, and arises from the unitarity of the scattering
S-matrix (which was not covered in this course).
pSemiclassically, we motivate this bound as follows. We write h̄bk = L '
h̄ `(` + 1) so that
` `+1
≤b≤ . (33.39)
k k
Assuming 100% scattering for particles with this range of b, the classical cross-
section would simply be the area of the annulus with radii k` and `+1
k :

π π
σ' ((` + 1)2 − `2 ) = 2 (2` + 1). (33.40)
k2 k
The actual unitarity bound is larger by a factor of 4. Again, the mysterious
quantum scattering factor of 4; recall the hard sphere cross section.

104
Tony Zhang 34 May 3, 2017 (recitation)

34 May 3, 2017 (recitation)


34.1 Wavepacket scattering
This section was heavily supplemented by Merzbacher 13.2 and Messiah 10
(good resources, according to a recitation instructor protip).
Consider an incident wavepacket

ψ(r, t = 0) = φ(r − r0 )

with impact parameter b, cross-sectional extent ∆r, and sharply defined wavevec-
tor k0 = k0 ẑ. We assume that r0 is sufficiently far from the origin such that
the entire wavepacket lies beyond the range a of the scattering potential. The
Fourier expansion of φ gives
Z
φ(r) = φ̃(k)eik·r d3 k.

where φ̃ is a smooth function of narrow extent ∆k centered with mean wavevector


k = k0 . We care what we see at a detector located distance D from the scattering
center and angle θ away from the ẑ axis. This distance should be very far away
from the center of the wavepacket (D sin θ  ∆r), so that it doesn’t pick up
the transmitted wave.
Now recall our results from plane wave scattering. With incident wave eik·r ,
we get a scattered wave ψks (r); the total wavefunction is just the sum of the
incident and scattered waves. The Fourier transform is an expansion in free-
particle eigenstates. But we want to expand in eigenstates of H so we know
how ψ evolves. For r → ∞, we know these eigenstates take the asymptotic form

eikr
ψk (r) = eik·r + fk (r̂) .
r
So for large r,
Z
ψ(r, t = 0) = φ̃(k)eik·(r−r0 ) d3 k
Z
' φ̃(k)ψk (r)e−ik·r0 d3 k
Z
φ̃(k) eik·r + ψks (r) e−ik·r0 d3 k.

=

We took our Fourier expansion (an expansion in free-particle eigenstates) and


promoted it to an expansion in H eigenstates, since the outgoing spherical
wave ψks makes very little contribution for large r.
Let us justify this last point better. We will show that in two regimes of
interest, r ∼ a and r → ∞, the outgoing waves ψks make negligible contribution.
SHOW WE CAN IGNORE IN r ∼ a.
mumble mumble r
D 1
, , a  ∆r  D
k0 k0

105
Tony Zhang 34 May 3, 2017 (recitation)

Let’s go to the large r case. Under time evolution for r → ∞,


Z
ψ(r, t) = φ̃(k)ψk (r)e−iEk t/h̄ e−ik·r0 d3 k

eikr
Z  
ih̄k2 t
= φ̃(k) e ik·r
+ fk (r̂) e− 2m e−ik·r0 d3 k
r
≡ φ(r, t) + ψs (r, t) (34.1)

We would like to approximate the dynamical phase in a way amenable to an


application of the inverse Fourier transform. Write
k 2 = k02 + 2(k − k0 ) · k + (k − k0 )2 .
The first term is just a constant, and the second term is amenable to an inverse
Fourier transform. We’d like to neglect the last term, which we should be able
to do, since we assumed that φ̃ is really sharply peaked around k0 . Let’s be
a little precise here. Since this expression contributes to a phase, we need the
resulting phase difference to be negligible:
h̄(k − k0 )2 t
 1. (34.2)
2m
If our detector lies r0 away from the scattering center, the scattered wave will
appear there at time
2mr0
T =
h̄k0
where the factor of 2 comes from the fact that it takes 12 T to get from r0 to the
origin and another 12 T to get back out. Plugging in t = T into Equation 34.2
and using (k − k0 )2 ∼ (∆k)2 , we find that we require
(∆k)2 r0
 1. (34.3)
k0
Physically, this condition enforces the fact that wavepacket dispersion at r0 is
negligible.
Assuming such is the case, the φ component in Equation 34.1 is therefore
Z
h̄k0
φ(r, t) ≈ eik·(r−r0 − m t) φ̃(k)eiω0 t d3 k

= φ(r − r0 − vt)eiω0 t
h̄k02
where ω0 = 2m . Our result makes sense, since the wave should travel at
h̄k0
velocity v = m . Now the scattered component looks like

eikr − ih̄k2 t −ik·r0 3


Z
ψs (r, t) = φ̃(k)fk (r̂) e 2m e d k
r
Z
fk (r̂) iω0 t ik·(k̂0 r−r0 −vt) 3
≈ φ̃(k) e e d k
r
fk0 (r̂) iω0 t
= e φ(k̂0 r − r0 − vt)
r
where we’ve again used the approximation for k 2 and where we’ve approximated
kr ≈ k0 · k̂0 r
(justified, again, because φ̃ is sharply peaked at k0 ).
TODO TODO reference chalkboard photo and Merzbacher 13

106
Tony Zhang 35 May 4, 2017

35 May 4, 2017
35.1 Scattering integral equation
Begin with the time-independent Schrödinger equation:

h̄2 2 h̄2 k 2
 
− ∇ + V (r) ψ(r) = Eψ(r) = ψ(r). (35.1)
2M 2M

If we define U by
h̄2
V (r) = U (r). (35.2)
2M
Then we arrive at an inhomogeneous linear ordinary differential equation

(∇2 + k 2 )ψ(r) = U (r)ψ(r). (35.3)

We introduce the Green’s function G, the solution to

(∇2 + k 2 )G(r − r0 ) = δ(r − r0 ). (35.4)

Intuitively, G is the function that produces the δ-function response when acted
upon by the differential operator on the left hand side; having G allows us to
determine solutions for any response function by simple integration.
We claim that our differential equation is equivalent to
Z
ψ(r) = ψ0 (r) + G(r − r0 )U (r0 )ψ(r0 ) d3 r (35.5)

where ψ0 solves the homogeneous differential equation (∇2 + k 2 )ψ0 (r) = 0 (i.e.
it’s a free-particle plane wave). Let us check our particular solution:
Z
(∇ + k )ψ(r) = (∇2 + k 2 )G(r − r0 ) U (r0 )ψ(r0 ) d3 r0
2 2
| {z }
δ(r−r0 )

= U (r)ψ(r)

as desired.
Moving on. Recall that we earlier showed that

eikr
(∇2 + k 2 ) =0 (35.6)
r
for r 6= 0. Moreover, we know that

∇2 (r−1 ) = −4πδ(r). (35.7)

Therefore, we guess that our Green’s function looks like

? 1 ±ikr
G± (r) = − e . (35.8)
4πr
Recall that G± must satisfy Equation 35.4.

107
Tony Zhang 35 May 4, 2017

Before we proceed to check it, we write down some helpful identities:

∇2 = ∇ · ∇
r
∇r =
r
∇ · (f v) = ∇f · v + f ∇ · v
r
∇f (r) = f 0 (r) .
r
Then we evaluate the Laplacian of G± :
  
2 2 ±ikr 1
∇ G± = ∇ e · −
4πr
   2  
1 1 1
= (∇2 e±ikr ) − + e±ikr ∇2 − + 2(∇e±ikr ) · ∇ −
4πr 4πr 4πr

Now we present the following useful results:


r
∇(e±ikr ) = ±ik e±ikr
 r 
2 ±ikr 2ik
∇ e = −k 2 ± e±ikr
r

so that, omitting a few final steps,

∇2 G± = −k 2 G± + δ(r)

as desired. Therefore, we’ve found two good Green’s functions. Which one do
we take in the our integral equation (Equation 35.5)? Note the sign on the
exponent in each. We want to model scattering. To get an outgoing wave in
Equation 35.5, we pick
0
1 e+ik|r−r |
G+ (r − r0 ) = − . (35.9)
4π |r − r0 |

Now we can also choose ψ0 (r) = eikz to get our incident wave. Therefore, write
the integral equation as
Z
ψ(r) = e + G+ (r − r0 )U (r0 )ψ(r0 ) d3 r0 .
ikz
(35.10)

Okeedokee. Those |r − r0 | things look really annoying to deal with. What


do we do? Since we’re usually really far from the origin with a detector, it’s
probably kosher to replace |r − r0 | 7→ r in the denominator. However, we can’t
do that in the numerator. Indeed, the wavelength 2π k could be comparable to
|r0 |, so we can’t make such a crude approximation in the numerator. Instead,
write
r = rn
so that to first order for r0  r,

|r − r0 | = r − n · r0

and
1 ikr −ikn·r0
G+ (r − r0 ) ≈ − e e . (35.11)
4πr
108
Tony Zhang 35 May 4, 2017

Plugging into Equation 35.10,


 Z  ikr
1 0 e
ψ(r) = eikz + − e−ikn·r U (r0 )ψ(r0 ) d3 r0 . (35.12)
4π r
Importantly, note that the quantity in brackets is now a function of n only: it
only has angular dependence now, so we’ve recovered our original scattering
wavefunction form!
Remark 35.1. Implicit in this approximation is the understanding that ev-
erywhere U is nontrivially large only for r0  r. That is, U decays sufficiently
quickly. We’ll see such is not the case for a Coulomb r−1 potential.

In general, we can write the incident wave as eikz 7→ eiki ·r . The outgoing
partial waves have ks ≡ nk. So we see that |ks | = |ki | = k. Then if the two k
are off by angle θ, their difference, the transfered momentum (well, wavenumber)
is
θ
K = |K| = 2k sin . (35.13)
2

35.2 The Born approximation


Before we get lost in the algebra, let’s get the spirit of the approximation. In
Equation 35.10, we hope that the integral term is smaller than the incident term.
In particular, if we have a high-energy particle incident on a weak potential, most
of the energy in ψ comes from eikz . Therefore, we might as well approximate
ψ ≈ eikz in the integral.
Indeed, let’s write
Z
ψ(r) = eikz + G+ (r − r0 )U (r0 )ψ(r0 ) d3 r0
Z
0
ψ(r ) = e ikz
+ G+ (r0 − r00 )U (r00 )ψ(r00 ) d3 r00 .

where we’ve just taken Equation 35.10 and added primes. Then substituting
the second into the first, we can write
Z
0
ψ(r) = eiki ·r + G+ (r − r0 )U (r0 )eiki ·r d3 r0
Z Z
+ d r G+ (r − r )U (r ) d3 r00 G+ (r0 − r00 )U (r00 )ψ(r00 ).
3 0 0 0

We could go on, adding more and more terms by iterative substitution, in the
spirit of Picard iteration. In general, we can expand into a full series solution
ψ(r) = ψ (0) (r) + ψ (1) (r) + . . .
where ψ (0) (r) = eiki ·r and where
Z
ψ (k+1)
(r) = G+ (r − r0 )U (r0 )ψ (k) (r0 ) d3 r0 .

Usually we just take the first two terms to get the first Born approximation.
Then Equation 35.12 becomes
 Z  ikr
iki ·r 1 −iks ·r0 0 iki ·r0 3 0 e
ψ(r) = e + − e U (r )e d r (35.14)
4π r
| {z }
fk (θ,φ)

109
Tony Zhang 35 May 4, 2017

where the partial wave coefficients are just


Z
1 0
fk (θ, φ) = − e−iK·r U (r0 ) d3 r0 , (35.15)

the Fourier transform of the potential evaluated at the transfer momentum
K = ks − ki . The transfer momentum encapsulates our knowledge of k, θ, φ.
How do we physically interpret the Born series? Well, ψ (0) just gives the
probability amplitude contribution from the incident wave. The next term
Z
ψ = G+ (r − r0 )U (r0 )ψ (0) (r0 ) d3 r0
(1)

can be interpreted as: taking the incident amplitude at each point in our
scattering potential, weighting by the potential, and propagating with G+ out to
our detector location r. (We’re convolving the amplitude contribution, weighted
by potential, with a term G+ that propagates things out to r.) The simple
picture is this contribution considers contributions from one wave deflection.
The second order term taking the contribution of the first order scattering,
weighting by potential, and propagating again, corresponding to two wave
deflections.

35.2.1 Radial potentials

If V (r) = V (r), we have


Z
1 2m 0
fk (θ) = − e−iK·r V (r0 ) d3 r0
4π h̄2
Z ∞ Z 1
m 02 0 0
=− 2πr dr d(cos θ)e−iKr cos θ
2πh̄2 0 −1
−iKr 0 0
m ∞ 02 0 − eiKr
Z
e
=− 2 r dr V (r0 )
h̄ 0 −iKr0
Z ∞
2m
=− rV (r) sin(Kr) dr (35.16)
Kh̄2 0
There are many more things we could mention about scattering, but for now,
know that Levinson’s theorem still holds, and that time delays still occur in
three dimensions.

110
Tony Zhang 36 May 8, 2017 (recitation)

36 May 8, 2017 (recitation)


36.1 Wavepacket scattering cross-section
Recall our discussion of wavepacket scattering. Today we’d like to find the
scattering cross-section
dNout
R ∞ dNout
dσ dt dΩ dt
= dN = R−∞
dt dΩ
∞ dNin
(36.1)
dΩ in dt
dt dA −∞ dt dA

Using outgoing probability current, we can write


dNout ~out .
= ~out · dS (36.2)
dt
But recall definition

~ = =(ψ ∗ ∇ψ) (36.3)
m
so that

=(ψs∗ ∇ψs )
~out = (36.4)
m
where we’ve invoked our scattered wave ψs .
Now we know that for large r,
f~k0 (Ω)
ψs (~r, t) = φ(k̂0 r − ~r0 − ~v t)eiω0 t (36.5)
r
Taking a gradient and neglecting O(r−2 ) terms,
f~k0 (Ω) d
∇ψs ≈ φ(k̂0 r − ~r0 − ~v t)eiωt r̂ (36.6)
r dr
f~ (Ω)
= k0 (∇φ(k̂0 r − ~r0 − ~v t) · k̂0 )eiωt r̂. (36.7)
r
This calls for a Fourier transform! We write
Z
~
φ(~r) = φ̃(k)eik·~r d3 k (36.8)
Z
~
∇φ(~r) = i~k φ̃(~k)eik·~r d3 k ≈ i~k0 φ(~r) (36.9)

where we assume φ̃ has narrow support (that is, our initial wavepacket had very
small momentum uncertainty).
We now find
dNout ~out = v|f~ (Ω)|2 |φ(k̂0 r − ~r0 − ~v t)|2 dΩ
= ~out · dS k0 (36.10)
dt
so the cross-section numerator becomes
Z ∞
dNout dNout
= dt (36.11)
dΩ −∞ dt dΩ
Z ∞
= v|f~k0 (Ω)|2 |φ(k̂0 r − ~r0 − ~v t)|2 dt (36.12)
−∞
Z ∞
2
= |f~k0 (Ω)| |φ(−bx , −by , ξ)|2 dξ (36.13)
−∞

111
Tony Zhang 36 May 8, 2017 (recitation)

where we now recall that ~r0 = (bx , by , −|z0 |) and substitute ξ = r + |z0 | − vt.
Finally, we want to integrate over all possible impact parameters, since we
imagine a beam of particles of varying impact parameter, distributed as P (~b):
Z
dNout dNout ~
= d2~bP (~b) (b) (36.14)
dΩ dΩ
Z Z ∞
= d2~bP (~b)|f~k0 (Ω)|2 |φ(−bx , −by , ξ)|2 dξ (36.15)
−∞
Z
≈ P (0)|f~k0 (Ω)|2 |φ(−bx , −by , ξ)|2 dξ d2 b (36.16)

where we assume that the outgoing differential flux dNout /dΩ is sharply peaked
around ~b = 0.
Yay, a numerator. Now we consider the denominator of the cross-section.
Write
dNin h̄
= (=ψ ∗ ∇ψ) · ẑ (36.17)
dt dA m

= =(φ∗ (~r, t)∇φ(~r, t)) · ẑ. (36.18)
m
Now with

φ(~r, t) = φ(~r − ~r0 − ~v t)eiω0 t (36.19)


∇φ(~r, t) = i~k0 φ(~r − ~r0 − ~v t)eiω0 t (36.20)

we get
dNin h̄k0
= |φ(~r − ~r0 − ~v t)|2 . (36.21)
dt dA m
Ingoing flux at (x, y, z) = (0, 0, −|z0 |), then, is
Z ∞
dNin dNin
(x = 0, y = 0, −|z0 |) = dt (36.22)
dA −∞ dt dA
Z ∞
=v |φ(−bx , −by , −vt)|2 dt (36.23)
−∞
Z ∞
= |φ(−bx , −by , ξ)|2 (36.24)
−∞

Again, we do the thing where we integrate out ~b, getting


Z
dNin
≈ P (0) |φ(−bx , −by , ξ)|2 dξ d2 b (36.25)
dA
so that

= |f~k0 (Ω)|2 . (36.26)
dΩ

36.2 A word on the Born aproximation


Recall the Born approximation,
Z
~ ~ 0
ψ(~r) = eiki ·~r + G+ (~r − ~r0 )U (~r0 )eiki ·~r d3 r0 + . . . (36.27)

112
Tony Zhang 36 May 8, 2017 (recitation)

where we have normalized potential


2m
U (~r) = V (~r) (36.28)
h̄2
and Green’s function
1 eik|~r|
G+ (~r) = − (36.29)
4π r
where k = |~ki |. For r → ∞,

~ f (Ω) ikr
ψ(~r) = eiki ·~r + e . (36.30)
r

Truncating the Born series to first order gives


Z
1 ~ ~ 0
f (Ω) = − U (~r0 )e−i(kf −ki )·~r d3 r0 . (36.31)

Write ~kf = kr̂. When is this approximation valid? Note that we have a series in
powers of U , so we want U to be small. But small compared to what?
Suppose the potential have depth V0 and spatial extent a. Then we want
the first term in the Born series small:
 
1 2m
a3 · · V0 1
a h̄2
or
h̄2
V0 
, (36.32)
ma2
giving the regime where the Born approximation is valid: the potential is weak
and there is no bound state.

113
Tony Zhang

Part VII

Identical particles

114
Tony Zhang 37 May 9, 2017

37 May 9, 2017
37.1 Identical particles
In quantum mechanics, we say that two particles are identical if they have the
same intrinsic properties (mass, spin, charge, magnetic moment, etc.). Their
other properties may differ (e.g. momentum, angular momentum).

37.1.1 Identical particles in classical mechanics

Let’s recall how we handle identical particles in classical mechanics: we assign a


labeling to the particles and follow them as the system evolves.

Example 37.1
If we have a particles at ~r0 and ~r00 with respective velocities ~v0 and ~v00 ,
Then we can write

~r1 (t0 ) = ~r0 ~r2 (t0 ) = ~r00


~v1 (t0 ) = ~v0 ~v2 (t0 ) = ~v00

But the labelling is arbitrary, so we should be able to specify the same


system as

~r1 (t0 ) = ~r00 ~r2 (t0 ) = ~r0


~v1 (t0 ) = ~v00 ~v2 (t0 ) = ~v0

Now suppose under the first labeling that we solve the system:

~r1 (t) = ~r(t) ~r2 (t) = ~r0 (t)

satisfying initial conditions

~r(t0 ) = ~r0 ~r0 (t0 ) = ~r00 .

Since we claim the particles are identical, the Hamiltonian must be invariant
under particle exchange:

H(~r1 , p~1 ; ~r2 , p~2 ) = H(~r2 , p~2 ; ~r1 , p~1 ). (37.1)

Of course, we could equally well write

~r1 (t) = ~r0 (t) ~r2 (t) = ~r(t).

The two labeling choices are equivalent, so in classical mechanics, pick one
of them and stick to your choice!

Now we move on to the quantum case. Two things change:

1. We can’t track the particles! Indeed, consider a two-particle scattering sys-


tem in the center-of-mass frame. We know how the wavepackets were sent
in, and we know where they end up. But we can’t determine which went

115
Tony Zhang 37 May 9, 2017

where, since their wavepackets were all superimposed near the scattering
center.
2. A more serious issue: exchange degeneracy. To illustrate, we consider
two spin- 12 particles in states |+i , |−i. Now with composite systems, we
work in the tensor product Hilbert space. Then we have basis states
|vi i(1) ⊗ |vj i(2) . (We can also write |vi i ⊗ |vj i, or |vi i |vj i, or yet shorter,
|vi , vj i.)
So our spin- 12 system is in state |+i(1) ⊗ |−i(2) or |−i(1) ⊗ |+i(2) . Since the
particles are identical, physically, the two states are the same. Moreover,
scaling a state or taking linear combinations with identical states gives
the same state, so the state

ψ = α |+i(1) ⊗ |−i(2) + β |−i(1) ⊗ |+i(2)

is equivalent physically. (Note, however, that the two basis kets here are
actually orthogonal!)
What is the probability to find the particles in the state

ψ0 = |+, xi(1) ⊗ |−, xi(2)


1   
= |+i(1) + |−i(2) ⊗ |+i(2) + |−i(2)
2
1
= (|++i + |+−i + |−+i + |−−i)
2
which gives p = |hψ|ψ0 i|2 = 14 |α + β|2 which depends on coefficients α and
β, bad.
With three particles, it gets worse. If the particles are in states φ, ω, χ, the
3! = 6 possible tensor products should all represent the same physical state.
In general, we will have N !-fold degeneracy for N identical particles.

Therefore, we introduce permutation operators, which are exactly what you


think they are. Consider the case of two distinguishable particles, where the
state spaces of the two are the same space V . Then given state |ui , uj i ∈ V ⊗ V ,
define P21 |ui , uj i = |uj , ui i.
2
Clearly, P21 = I = P12 . None of this stuff should be surprising, because P12
and P21 form a representation of S2 on V ⊗2 .
We claim that P21 is Hermitian:

(1) huk | ⊗ (2) hu` |P21 |ui i(1) ⊗ |uj i(2) = δkj δ`i
 ∗
† †
(1) huk | ⊗ (2) hu ` |P 21 |ui i(1) ⊗ |uj i(2) = (1) hui | ⊗ (2) hu j |P 21 |u k i (1) ⊗ |u` i(2)

= δ`i δkj

Indeed, P21 is unitary!


Now we say states are

1. symmetric if P21 |ψS i = |ψS i


2. antisymmetric if P21 |ψA i = − |ψA i

116
Tony Zhang 37 May 9, 2017

Much like we can decompose all functions into odd and even functions, we
can decompose V ⊗ V = Sym2 V ⊕ Anti2 V . We introduce Hermitian projec-
tion operators S and A that (anti)symmetrize states (i.e. project onto the
(anti)symmetric subspace):
1
S= (1 + P21 ) (37.2)
2
1
A = (1 − P21 ) (37.3)
2
(From here, we refer only to Hermitian projection operators.) It is straight-
forward to verify that S 2 = S and A2 = A. Furthermore, S + A = 1, so the
projectors are complementary: they resolve the identity (albeit over a direct
sum decomposition instead of a basis).
Using the facts (easily checked) that P21 S = SP21 = S and P21 A = AP21 =
−A, it should be clear that S |ψi is symmetric and A |ψi is antisymmetric.
Before generalizing to more particles, we consider the action of S2 on opera-
tors. Let B ∈ Hom(V ) and define B(1) = B ⊗ 1 (and B(2) = 1 ⊗ B).
Then
 

P21 B(1)P21 |ui i(1) ⊗ |uj i(2) = P21 B(1) |uj i(1) ⊗ |ui i(2)
= P21 (B |uj i(1) ) ⊗ |u1 i(2)
= |ui i(1) ⊗ (B |uj i(2) )
 
= B(2) |ui i(1) ⊗ |uj i(2)

Here’s a sneak peak of the N -particle case: for each of the N ! permutations
σ ∈ SN , we have an operator Pσ (e.g. P231 ) that does an obvious thing:

P231 |ui i(1) ⊗ |uj i(2) ⊗ |uk i(3) = |uk i(1) ⊗ |ui i(2) ⊗ |uj i(3) .

Basically we have a representation of SN on V ⊗n .

117
Tony Zhang 38 May 10, 2017 (recitation)

38 May 10, 2017 (recitation)


38.1 More on the Born approximation
This discussion covers material from Landau and Lifshitz, sections 125 and 45.
Given incident plane wave eiki ·r , the Born approximation gives, to first order,
Z
0
iki ·r
ψ(r) = e + d3 r0 G+ (r − r0 )U (r0 )eiki ·r + O(u2 ) (38.1)

where we have dimensionless potential


2m
U= V (38.2)
h̄2
with Green’s function
eikr
G+ (r) = − . (38.3)
r
If we suppose U has maximum depth U0 and range a, we can treat U as a
h̄2
perturbation if ka . 1 or if U0 a2  1 (i.e when V0  ma 2 ).

TODO TODO TODO see board pictures


It turns out that in one dimension, as k → 0, the Born approximation always
1 ik|x|
fails. Indeed, G+ (x) = 2ik e ; this term will diverge while the other terms in
the first-order Born summand. Indeed, we must have
aU0
 1, (38.4)
k
which gives
TODO TODO see picture

118
Tony Zhang 39 May 11, 2017

39 May 11, 2017


39.1 The symmetric group
Last time we started follow Cohen-Tannoudji’s notation for permutations, which
is opposite the usual mathematical notation. Today, we’ll switch to the conven-
tional notation.
Recall last time that we defined P21 acting on V ⊗ V by

P21 |ui i(1) ⊗ |uj i(2) = |uj i(1) ⊗ |ui i(2) .

Now recall how we define P231 last time. Expunge that from memory. We’re
now going to define
P231 |ui uj uk i = |uj uk ui i
where we use the shorthand |abc . . .i = |ai(1) ⊗ |bi(2) ⊗ . . . .
What is the inverse of P231 ? Why, it’s simply P312 , corresponding to the
inverse of the original permutation in S3 .
In general, we can consider general permutation operators Pα where α ∈ SN
is a permutation of {1, . . . , N }. We can think of it as a list {α(1), . . . , α(N )}.
They act as
Pα |u1 u2 . . . uN i = |uα(1) uα(2) . . . uα(N ) i . (39.1)

Example 39.1
P3142 |u1 u2 u3 u4 i = |u3 u1 u4 u2 i .

How do we compose permutation operators? Let’s look at an example:

Example 39.2
Heavily abbreviating,

P4123 P3214 |abcdi = P4123 |cbadi = |dcbai = P4321 |abcdi .

Note that this composition is the reverse of what you might expect from
group theory. Indeed, we’re using the law of composition στ = τ ◦ σ: we’re
using right multiplication.

What followed was some group theory discussion:

• Groups have associative multiplication with identity and inverse.


• The symmetric group SN is the set of permutations of N things under
composition; it has order N !.
• While S2 is trivial, S3 has a more interesting structure. There’s an identity,
three transpositions, and two cycles. (That is, the class equation of S3
is more interesting, with cycle structures that aren’t the identity or a
transposition.)
• The transpositions in SN generate the whole group. Professor introduces
cycle notation (well, the only cycles are transpositions).

119
Tony Zhang 39 May 11, 2017

• We can define the sign of a permutation generated by a product of


k transpositions as (−1)k . The sign is well-defined. The parity of a
permutation is just the parity of k.
• There are 12 N ! even (resp. odd) permutations. To see, the sign function is
a surjective homomorphism SN → {±1}, and the even permutations form
the kernel. (In class: ad hoc proof by considering the even/odd cosets,
then bijecting them by action of a fixed transposition.)

Now recall that P21 is Hermitian (by inner-product-bash) and unitary (be-
2
cause P12 = 1). By the same argument, Pt is unitary for any transposition t.
Since transpositions generate SN , any Pα is unitary. However, they’re not all

Hermitian! (For example, consider P312 .)
Joke. (On the group multiplication table for S3 ) “It’s like sudoku!” –zwiebach

39.2 Symmetrizers and antisymmetrizers


Can we find a simultaneous eigenstate of all N ! permutation operators Pα ? Since
they’re not all Hermitian and not commuting, we won’t have a complete set of
simultaneous eigenstates. But we can still find some simultaneous eigenstates.
For example, symmetric states |ψS i, invariant under action by all Pα , are
simultaneous eigenstates of all Pα with eigenvalue 1.
Similarly, antisymmetric states |ψA i, satisfying Pα |ψA i = α |ψA i (where α
is the sign of α), are simultaneous eigenstates of all Pα .
Remark 39.3. These two classes of states correspond to two one-dimensional
representations of SN . The optional pset will look at the N = 3 case (in
particular, it will look at the other irreducible representation of S3 ).

Now we can define SymN V (resp. AntiN V ) as the subspaces of symmet-


ric (resp. antisymmetric) states in V ⊗N . Furthermore, we can define the
symmetrizer S and the antisymmetrizer A:
1 X
S= Pα (39.2)
N!
α∈SN
1 X
A=  α Pα . (39.3)
N!
α∈SN

Since taking adjoints permutes the Pα , S is Hermitian. Also, taking adjoints of


Pα (i.e. inverses, because Pα is unitary) doesn’t affect the sign of P α, so A is
Hermitian as well.
We have some more identities:

Pα S = SPα = S (39.4)
Pα A = APα = α A (39.5)

which can be proven easily.


Now S and A are actually Hermitian projectors:

S2 = S A2 = A SA = AS. (39.6)

onto orthogonal subspaces.

120
Tony Zhang 39 May 11, 2017

Remark 39.4. Importantly, an idempotent operator is a projector in the sense


that its kernel and image are orthogonal iff it is Hermitian as well.

These identities show that S (resp. A) sends states to (anti)symmetric states.

121
Tony Zhang 40 May 15, 2017 (recitation)

40 May 15, 2017 (recitation)


40.1 About identical particles
Recall our setup. We have N identical particles whose individual state spaces
are V . Then the N -particle state lives in V ⊗N . For any permutation α ∈ SN ,
let Pα be the operator that acts on product states as

Pα |a1 , a2 , . . . , aN i = |aα(1) , . . . , aα(N ) i .

We say a state is symmetric if Pα |ψi = |ψi for all α; it’s antisymmetric if


Pα |ψi = sgn(α) |ψi.
Now for identical particles, we demand that permutations of particles give
the same element of V ⊗N , up to phase. If α is a transposition, then since Pα2 is
the identity, Pα |ψi = ± |ψi.
We say that a particle is a boson if transposition Pα gives the same state.
Then any permutation produces the same state, and so boson states are symmet-
ric. Conversely, a particle is a fermion if interchange flips the sign. So fermion
states are antisymmetric.
Introduce symmetrizer and antisymmetrizer
1 X 1 X
S= Pα A= sgn(α) Pα . (40.1)
N! N! | {z }
α∈SN α∈SN εα

We claim they do what their names say they do:

Pα S = SPα = S Pα A = APα = εα A. (40.2)

Basically, S makes states symmetric; A, antisymmetric. From these properties, it


follows that these operators are projectors: indeed, they’re idempotent (S 2 = S
and A2 = A). They project onto orthogonal subspaces: SA = AS = 0.
Proving all of these requires the simple observation that P is a representation
of SN on V ⊗N . Indeed
Pα Pβ = Pα·β
where we define α · β = β ◦ α. (This definition is the reverse of the usual; it’s
required because of the way we defined Pα , which was right-multiplicative.)
The actual proofs are straightforward group theory exercises, so I’m not
going to copy them here.

122
Tony Zhang 40 May 15, 2017 (recitation)

Example 40.1
Consider two identical spin- 12 particles. So our Hilbert space is V ⊗ V ,
where V ∼= C2 . Acting on the product space basis with S gives three
symmetric states

S |↑↑i = |↑↑i
|↑↓i + |↑↓i
S |↑↓i = S |↓↑i =
2
S |↓↓i = |↓↓i

corresponding to the triplet states.


Now

A |↑↑i = 0
|↑↓i − |↓↑i
A |↑↓i =
2
|↓↑i − |↑↓i
A |↓↑i =
2
A |↓↓i = 0

If we have fermions, we see that we cannot have both particles in the same
spin state. This is basically the Pauli exclusion principle.

123
Tony Zhang 41 May 16, 2017

41 May 16, 2017


41.1 The symmetrization postulate
In a system with N identical particles, all physical states belong
either to SymN V (in which case the particles are bosons) or to
AntiN V (resp. fermions).

Some comments:

1. This claim is an additional postulate in quantum mechanics. It’s true in


three dimensions, but there are more options in two dimensions (anyons).

2. Symmetrization determines the statistical properties of bosons and fermions.


3. The spin-statistics theorem from quantum field theory tells us that bosons
(resp. fermions) have integer (resp. half-integer) spin.
4. Suppose we have a composite system of bosons and fermions. (For con-
creteness, consider a hydrogen atom: an electron bound to a proton.) If
we have multiple such systems, what happens to the wavefunction under
exchange? The parity is just the sum of the parities of the constituents! In
the case of two hydrogen atoms, both protons and electrons are fermions,
so
ψ(e2 , p2 , e1 , p1 ) = −ψ(e1 , p2 , e2 , p1 ) = ψ(e1 , p1 , e2 , p2 ).

5. The symmetrization postulate resolves the exchange degeneracy problem!


Let |ui ∈ V ⊗N . Then let V|ui be the span of all Pα |ui, and say that the
exchange degeneracy of |ui is dim V|ui , which ranges from 1 to N !.
In fact, we claim that V|ui ∩ SymN V and V|ui ∩ AntiN V are both one
dimensional: there is a unique symmetric and unique antisymmetric vector
in V|ui up to scale factor.
Take |ψi ∈ V|ui ∩ SymN V . Expand it in the Pα |ui basis:
X
|ψi = cα Pα |ui .
α

Then
!
X X X
|ψi = S |ψi = cα SPα |ui = cα S |ui = S |ui cα
α α α

which is parallel to S |ui, which we now see gives the unique symmetric
subspace of V|ui .
Antisymmetrization is a little more involved.

41.2 Building antisymmetric states


How can we antisymmetrize a state? For concreteness, consider three-particle
state
|ui = |ϕi(1) ⊗ |χi(2) ⊗ |ωi(3) .

124
Tony Zhang 41 May 16, 2017

Then in fact we can write


|ϕi(1) |ϕi(2) |ϕi(3)


1 X 1
A |ui = α Pα |ui = |χi(1) |χi(2) |χi(3)
3! α 3!
|ωi(1) |ωi(2) |ωi(3)

It’s important that we keep the labels on the kets: we might get a term like
− |ϕi(2) ⊗ |χi(1) ⊗ |ωi(3) , which is really the state − |χϕωi.
It’s nice how the signs work out with the determinant. In fact, the definition
of A |ui matches that of the determinant

X N
Y
det B = α Bα(i),i , (41.1)
α i=1

so we can actually use this determinant for N particle systems.


Then defining ωij = |ωi i(j) , we can write
N
O O
|ωi = |ωi i(i) = ωii (41.2)
i=1 i
N
O O
Pα |ωi = |ωα(i) i(i) = ωα(i),i (41.3)
i=1 i

whence
1 X 1 X O 1
A |ωi = α Pα |ωi = α ωα(i),i = det ωij .
N! α N! α i
N !

This determinant is called the Slater determinant. It’s clear that we can’t
have two |ωi i the same, because then the determinant vanishes. So two fermions
cannot occupy the same state: this is the Pauli exclusion principle!

41.3 Occupation numbers


After introducing this edifice of permutation operators, symmetrization and
antisymmetrization, and Slater determinants, we’ve solved the problem of
exchange degeneracy. Yay. But now specifying a state is hideous and sad.
As before, suppose we have N identical particles, so our Hilbert space is V ⊗N .
Suppose we have an orthonormal basis |ui i (possibly infinite) of V . Consider
the basis state
ON
|uki i(i)
k=1
⊗N
in V . To obtain the corresponding physical state, we (anti)symmetrize.
Define occupation number nk as the number of times |uk i appears in the
tensor product basis state. It’s clear that if two tensor product basis states have
the same occupation numbers, (anti)symmetrization produces the same state.
Conversely, we can get the same state upon (anti)symmetrization of two tensor
product basis states only if they share the same occupation numbers.
So for bosons, we can just work with states
⊗n
O
|n1 , n2 , . . .i ∝ S |uk i k .
k

125
Tony Zhang 41 May 16, 2017
P
The tensor product is finite, since k nk = N . These states are all orthogonal
too!
We summarize our discussion with the following:

Proposition 41.1
• For N bosons, the states in SymN V are |n1 n2 . . .i with nonnegative
integer occupation numbers ni summing to N .

• For N fermions, the states in AntiN V are |n1 n2 . . .i with i ni = N


P
and ni ∈ {0, 1}.

41.4 Particles in product spaces


Now we ask: suppose each identical particle has tensor product state space
V ⊗ W (e.g. electron position and electron spin)?
There’s not much time left, so we work out an example for N = 2:

Sym2 (V ⊗ W ) = (Sym2 V ⊗ Sym2 W ) ⊕ (Anti2 V ⊗ Anti2 W )


Anti2 (V ⊗ W ) = (Sym2 V ⊗ Anti2 W ) ⊕ (Anti2 V ⊗ Sym2 V )

In general, defining

|γabi = · · · ⊗ |ai ⊗ · · · ⊗ |bi ⊗ · · · , (41.4)

we can write
1 1
|γabi = (|γabi + |γbai) + (|γabi − |γbai) (41.5)
2 2

126
Tony Zhang 43 May 17, 2017 (recitation)

42 Exchange forces (home reading)


TODOO read about exchange forces in Griffiths

43 May 17, 2017 (recitation)


TOD TOD TODO

127
Tony Zhang 44 May 18, 2017

44 May 18, 2017


44.1 Identical composite systems
We continue our discussion of systems of indistinguishable particles where
single particles have tensor product state spaces. Suppose a particle has state
space V ⊗ W . We want to consider the two-identical-particle wavefunctions.
For concreteness, if we consider two electrons, we might have a normalized
wavefunction
ψ(x1 , m1 ; x2 , m2 ) = φ(x1 , x2 )χ(m1 , m2 ). (44.1)
where m is the Sz spin quantum number. The probability to find an electron
within d3 x1 of x1 and d3 x2 of x2 is then given by
dp = |φ(x1 , x2 )|2 d3 x1 d3 x2 . (44.2)

Suppose the electrons are non-interacting, so that we can write a Hamiltonian


h̄2 2 h̄2 2
 
− ∇1 + V (x1 ) − ∇2 + V (x2 ) ψ = Eψ. (44.3)
2m 2m
This Schrödinger equation is separable, yielding a family of solutions of the
form ψA (x1 )ψB (x2 ). Now to yield a antisymmetric total wavefunction ψ, spatial
wavefunction φ must be symmetric or antisymmetric. Thus, consider spatial
solutions
1
φ± (x1 , x2 ) = √ (ψA (x1 )ψB (x2 ) ± ψA (x2 )ψB (x2 )). (44.4)
2
If we take φ+ (resp. φ− ), the total wavefunction will be φ+ χs (resp. φ− χt ),
where χs is the singlet spin state and χt is any one of the triplet spin states.
We calculate the spatial probability density:
dp± = |φ± (x1 , x2 )|2 d3 x1 d3 x2
1
= (|ψA (x1 )|2 |ψB (x2 )|2 + |ψA (x2 )|2 |ψB (x1 )|2
2
∗ ∗
± 2<(ψA (x1 )ψA (x2 )ψB (x2 )ψB (x1 ))) d3 x1 d3 x2
In the special case where x1 = x2 = x,
(
3 3 2|ψA (x1 )|2 |ψB (x2 )|2 + case, singlet
dp± = d x1 d x2 (44.5)
0 − case, triplet
which makes sense: if we’re spatially symmetric, the spins are antisymmetric,
so the probability of finding two electrons at x is twice what we might expect
semiclassically. If we’re spatially antisymmetric, there is no probability of finding
both electrons in the same spot, which makes sense, since the spins are symmetric
(read: parallel) now.
Now we perform a sanity check: a sort of “antisymmetrizing over all elec-
trons”. Assume ψA and ψB have disjoint supports RA and RB , and consider
the probability that we find particles at x1 ∈ RA and x2 ∈ RB . Again,
1 0 0
(|ψA (x1 )|2 |ψB (x2 )|2 +  2  2
 : :
dp± = |ψ
A (x2 )| |ψ
B (x1 )|
2 
∗ : 0ψ ∗ (x ) ψ  :0 3
± 2<(ψA (x1 )ψA(x
2)

B 2 B (x1 )
 )) d x1 d3 x2
so we get no interference between two separated particles. Therefore, if two
particles have disjoint support, we don’t have to antisymmetrize over them.

128
Tony Zhang 44 May 18, 2017

44.2 Counting states and distributions


Suppose we have energy levels Ei with degeneracy di for a single particle.
Further, suppose we have N particles, with energy level occupation numbers Ni .
Then we can define Q(N1 , . . . ) as the number of ways to distribute N particles
so that Nj live in level j. The form of Q will depend on the distinguishability
of the particles (and if not, whether they’re fermions or bosons). As you can
see, we’re about to have some fun times with combinatorics.

44.2.1 Maxwell-Boltzmann statistics (distinguishable particles)

In this case, Q is just the number of ways to distribute N distinguishable balls


into distinguishable urns with Ni in each urn i, such that all balls in urn i are
labeled with one of di labels:
Y (di )Ni
QMB = N ! . (44.6)
i
Ni !

44.2.2 Fermi-Dirac statistics (identical fermions)

For identical particles, specifying the occupation numbers Ni gives the state
uniquely. So it remains to consider the number of ways we can arrange particles
within the degenerate subspaces. For subspace i, there are
 
di di !
= (44.7)
Ni Ni ! (di − Ni )!

possibilities, since the exclusion principle forbids multiple occupation. Therefore,


Y  di 
QFD = . (44.8)
i
Ni

44.2.3 Bose-Einstein statistics (identical bosons)

Again, we want to find the number of ways to arrange Ni indistinguishable


particles in di distinguishable slots where we now allow multiple occuapation of
slots. This is just a stars and bars problem, and the answer is just
 
di + Ni − 1 (di + Ni − 1)!
= (44.9)
Ni Ni ! (di − 1)!

so that
Y di + Ni − 1
QBE = . (44.10)
i
Ni

44.2.4 Deriving distributions

We now want to find the most probable (i.e. entropy-maximizing) configura-


tion {N1 , N2 , . . . } under Fermi-Dirac or Bose-Einstein statistics, assuming fixed
energy and particle number. But the likelihood of a configuration is proportional
to Q(N1 , . . . ), since the fundamental assumption of statistical mechanics

129
Tony Zhang 44 May 18, 2017

tells us that all states with equal total energy are equally likely when we’re in
thermal equilibrium.
P
P Therefore, we maximize Q subject to constraints N = i Ni and E =
i N i Ei . It’s actually more convenient to maximize log-likelihood ln Q, which
we’ll do with Lagrange multipliers. Define Lagrangian
! !
X X
f (N1 , . . . ; α, β) = ln Q(N1 , . . . ) + α N − Ni + β E − Ei Ni .
i i
(44.11)
We’ll be considering factorials with large arguments, so we use Stirling’s
approximation
√  n n
n! ∼ 2πn , (44.12)
e
or, taking logs,
ln n! ∼ n ln n − n. (44.13)
Then the log-likelihood in the Fermi-Dirac case is
X di !
ln QFD = ln
i
Ni ! (di − Ni )!
X
= (ln di ! − ln Ni ! − ln(di − Ni )! )
i
X
≈ di ln di − Ni ln Ni − (di − Ni ) ln(di − Ni )
i

Therefore, we can differentiate the Lagrangian as


∂f
0= = − ln Ni − 1 + ln(di − Ni ) + 1 − α − βEi
∂Ni
di − Ni
= ln − α − βEi
Ni
As a result,
di
Ni = α+βE
. (44.14)
e i +1
It turns out that α and β are, with chemical potential µ(T ),
µ(T ) 1
α=− β= (44.15)
kB T kB T
so that
di
Ni = .
Ei −µ(T )
(44.16)
e +1kB T

We arrive at the Fermi-Dirac distribution, which gives the average number of


particles in a single state:
1
ni = E−µ . (44.17)
e kB T
+1
Similar manipulations give the Maxwell-Boltzmann distribution
di
ni = E−µ (44.18)
e kB T

and the Bose-Einstein distribution


di
ni = E−µ . (44.19)
e kB T
−1

130
Tony Zhang A Useful formulae

A Useful formulae
A.1 Wave mechanics
Probability current

~ = =(ψ ∗ ∇ψ) (A.1)
m

A.2 Spin- 12

Spin operators are given by Si = 12 h̄σi , where we have Pauli matrices


     
0 1 0 −i 1 0
σx = σy = σz = . (A.2)
1 0 i 0 0 −1

These obey

σa σb = δab I + iεabc σc [σa , σb ] = 2iεabc σc {σa , σb } = 2δab I. (A.3)

An arbitrary 2 × 2 Hermitian matrix a0 I + ~a · ~σ has eigenvalues

λ = a0 ± |~a|. (A.4)

A.3 Harmonic oscillator


Hamiltonian:
p̂2
   
1 2 2 † 1 1
Ĥ = + mω x̂ = h̄ω â â + = h̄ω N̂ + (A.5)
2m 2 2 2

with characteristic length and momentum


r
2h̄
x0 = p0 = 2h̄mω, (A.6)

creation and annihilation operators
x̂ p̂ x̂ p̂
â = +i ↠= −i (A.7)
x0 p0 x0 p0
and number operator
N̂ = ↠â. (A.8)

The spectrum is parameterized by nonnegative integer n, with energy levels


and eigenstates
n
â†
 
1
En = h̄ω n + |ni = √ |0i (A.9)
2 n!
where |0i is the ground state. Raising and lowering operators act as
√ √
↠|ni = n + 1 |n + 1i â |ni = n |n − 1i (A.10)

131
Tony Zhang A Useful formulae

A.4 Angular momentum


The angular momentum algebra consists of three operators J1 , J2 , J3 that gen-
erate some sort of rotation. In particular, the span of the operators Ji /ih̄ is
isomorphic to the Lie algebra so(3):

[Ji , Jj ] = ih̄εijk Jk . (A.11)

We can define J 2 = Ji Ji and ladder operators:

J± = Jx ± iJy . (A.12)

For a given multiplet with fixed j quantum number, we have


p
J± |j, mi = h̄ j(j + 1) − m(m ± 1) |j, m ± 1i (A.13)
p
= h̄ (j ± m + 1)(j ∓ m) |j, m ± 1i (A.14)

Also,

J± J∓ = J12 + J22 ± h̄J3 J 2 = J+ J− + J3 (J3 − h̄). (A.15)

A.5 Hydrogen atom


We use Gaussian cgs units throughout:

4πε0 = 1. (A.16)
B
and magnetic fields should be converted from SI as B 7→ c.

A.5.1 Gross structure

Hamiltonian with Coulomb potential:

p̂2 e2 h̄2 2 e2
Ĥ = − =− ∇ − . (A.17)
2me r̂ 2me r
Bohr radius:
h̄2 h̄
a0 = = = 52.918 pm. (A.18)
me e2 αmc
Energy levels:
e2 1 α2 mc2 13.6 eV
En = − = − =− . (A.19)
2a0 n2 2 n2 n2
Simultaneous Ĥ, L̂2 , L̂z eigenstates:
s 3
2 (n − ` − 1)! − 1 ρ ` 2`+1
ψn`m (r, θ, φ) = e 2 ρ Ln−`−1 (ρ)Y`m (θ, φ) (A.20)
na0 2n(n + `)!

where we define a dimensionless length scale


2r
ρ= , (A.21)
na0

where L2`+1
n−`−1 is a generalized Laguerre polynomial of degree n − ` − 1, and
where Y`m is a spherical harmonic (orthonormalized, with the Condon-Shortley
phase convention).

132
Tony Zhang A Useful formulae

Ground state:
1 r
ψ100 (r, θ, φ) = p 3 e− a0 (A.22)
πa0
n = 2 states:
 
1 r r
ψ200 = p 1 − e− 2a0 (A.23)
8πa03 2a0
1 ∓x − iy − 2ar
ψ21,±1 = p 3 e 0 (A.24)
8 πa0 a0
1 z − 2ar
ψ210 = p e 0 (A.25)
4 2πa0 a0
3

A.5.2 Fine structure

Fine structure constant:


e2 1

α= (A.26)
h̄c 137
Hydrogen atom Hamiltonian with fine-structure perturbation:

p2 p4 1 1 dV h̄2
H= +V − 3 2
+ 2 2
S · L + 2 2
∇2 V (A.27)
|2m{z } |8m{zc } |2m c r{zdr } |8m c{z }
H0 δHrel δHspin-orbit δHDarwin

Fine structure energy shifts:

1 α4 mc2
 
(1) n 3
δEn,j = − 1 − (A.28)
2 n4 j+ 2
4

A.5.3 Zeeman effect

Zeeman effect perturbation:


e
δH = B · (L + 2S) (A.29)
2mc
The weak and strong cases depend on the ordering of eh̄B 4 2
2mc and α mc . In
the weak case, the B-field perturbs the fine-structure levels; in the strong,
fine-structure interactions perturb the strong field levels.

A.6 Time-independent perturbation theory


A.6.1 Nondegenerate case

First order energy shift:


En(1) = hn(0) |δH|n(0) i (A.30)
kth order:
En(k) = hn(0) |δH|n(k−1) i . (A.31)

To determine wavefunction shifts and higher-order energy shifts, use

(H 0 −En(0) ) |n(k) i = (En(1) −δH) |n(k−1) i+En(2) |n(k−2) i+· · ·+En(k) |n(0) i (A.32)

133
Tony Zhang A Useful formulae

and resolve the identity as necessary. Most useful is the first order wavefunction
shift:
X δHkn
|n(1) i = − (0) (0)
|k (0) i (A.33)
E
k6=n k − En

which gives second order energy shift


X |δHnk |2
En(2) = (0) (0)
. (A.34)
k6=n En − Ek

A.6.2 Degenerate case

For valid first order results, we must choose a “good basis” in which we diago-
nalize δH in degenerate subspace n:
(1)
(δH)nl,nk = Enk δlk . (A.35)
A sufficient condition for the goodness of a basis is the existence of a complete
set of commuting observables on our degenerate subspace with our basis as
common eigenbasis that also commutes with δH.
(1)
If all the Enk are different, degeneracy is broken to first order, and the
first-order wavefunction shifts are
X δHp,nk XX δHnl,p δHp,nk
|n(1) ; ki = (0) (0)
+ (0) (0) (1) (1)
|n(0) ; li
p6=n En − Ep l6=k p6=n (En − Ep )(Enk − Enl )
(A.36)
The second order energy shifts will come easily.
Otherwise, our “good basis” is only good to first order. To get second order
energy shifts and first order wavefunction shifts, we’ll need to go to second order.
A basis good to second order must diagonalize
(2)
X δHnl,p δHp,nk
Mlk = (0) (0)
. (A.37)
p6=n En − Ep
(2)
The resulting eigenvalues are the En .

A.7 WKB approximation


For a one-dimensional free particle Hamiltonian
p2
H= + V (x),
2m
we can define
p p
2m(E − V (x)) 2m(V (x) − E)
k(x) = κ(x) = (A.38)
h̄ h̄
The semiclassical approximation gives solutions of the form
A Rx
k(x0 ) dx0 B Rx
k(x0 ) dx0
ψ(x) = p ei a +p e−i a (A.39)
k(x) k(x)
A Rx
κ(x0 ) dx0 B Rx 0
ψ(x) = p ei a +p e− a κ(x ) dx (A.40)
κ(x) κ(x)

134
Tony Zhang A Useful formulae

for classically allowed and classically forbidden regions, respectively. This


approximation is good when the following equivalent conditions holds

dλ dp
λ  λ λ  |p| (A.41)
dx dx

where λ is the local de Broglie wavelength.


If we move from allowed to forbidden at x = a:
 Ra Ra
 √2A cos x k(x0 ) dx0 − π4 − √B sin x k(x0 ) dx0 − π4
 
xa
k(x) k(x)
ψ(x) =
 √ A exp − ax κ(x0 ) dx0 + √ B exp ax κ(x0 ) dx0
R  R 
xa
κ(x) κ(x)
(A.42)
Moving from forbidden to allowed at x = b:
  R  R 
 √ A exp − xb κ(x0 ) dx0 + √ B exp xb κ(x0 ) dx0 x<b
κ(x) κ(x)
ψ(x) =
 √2A cos x k(x0 ) dx0 − π − √B sin x k(x0 ) dx0 − π
R  R 
b 4 b 4 x>b
k(x) k(x)
(A.43)
We can connect away (i.e. toward a turning point) from a decaying exponential
and into a growing exponential.
In the special case of quantum tunneling through a barrier, the probability
suppression is !
Z b
0 0
T = exp −2 κ(x ) dx (A.44)
a

for turning points a < b.

A.8 Time-dependent perturbation theory


Given time-dependent perturbation δH(t) to time-independent Hamiltonian H0 ,
we move to the interaction picture. Use unitary time evolution operator U =
iH0 t
e− h̄ to cancel out trivial time dependence from H0 to get

|ψ̃(t)i = U † |ψ(t)i δH(t)


f = U † (t)δHU (t) (A.45)

and

|ψ̃(t)i = δH(t)
ih̄ f |ψ̃(t)i . (A.46)
∂t
The perturbative solution, to second order, is just

t0
!
Z t Z t Z
1 0 f 0) + 1 0 00 0 f 00
|ψ̃(t)i = 1+ dt δH(t dt dt δH(t )δH(t ) |ψ̃(0)i
f
ih̄ 0 (ih̄)2 0 0
(A.47)

with obvious generalizations to higher order.


We can expand |ψ̃(t)i in the spectrum of H0 with time-dependent coefficients
and solve for them too.

135
Tony Zhang A Useful formulae

A.8.1 Fermi’s Golden Rule

With a time-independent perturbation H 0 turning on at t = 0 or a harmonic


perturbation 2H 0 cos ωt, the transition rate from state i into continuum state f
is

Γf ←i = ρ(Ef )|hf |H 0 |ii|2 (A.48)

where

• ρ(E) gives the state density dN


dE where N is the number of energy eigenstates
with energy E. You may have to regularize by making the universe a large,
but bounded, box.

• Ef = Ei in the constant case or Ef = Ei ± h̄ω in the harmonic case


• Valid for times 2πh̄ 1
∆  t  Γ , where ∆ is an energy range sufficiently
small such that ρ and Vf i are both effectively constant

A.8.2 Emission rates

Einstein A and B coefficients:


3
4 ωba ~ 2 4π 2 ~ 2
A= |ha|d|bi| B= |ha|d|bi| (A.49)
3 h̄c3 3h̄2

where d~ is the dipole moment operator. These coefficients give the rates of
spontaneous emission and of stimulated emission per unit spectral energy density,
respectively, for an atom subject to thermal radiation, with spectral energy
density given by Equation A.52.
The actual absorption rate is Bab U (ωab )Na ; stimulated emission, Bab U (ωab )Nb ;
spontaneous emission, ANb . Use the fact that Na ∝ e−βEa .

A.8.3 Selection rules for hydrogen

The only possible electric dipole transitions in gross-structure hydrogen have

`0 − ` = ±1 m0 − m = −1, 0, 1. (A.50)

A.9 Thermodynamics
Thermodynamic beta:
1
β= . (A.51)
kT
Spectral energy density of blackbody radiation,

ω3
 

U (ω) = 2 3 . (A.52)
π c eβh̄ω − 1

(So that U (ω)dω dV is the energy in volume dV in range dω.)

136
Tony Zhang A Useful formulae

A.10 Adiabatic approximation


Given slow-evolving time-dependent Hamiltonian H(t) with instantaneous spec-
trum
H(t) |ψn (t)i = En (t) |ψn (t)i , (A.53)
a system beginning in |ψn (0)i evolves as
|ψ(t)i ' eiθn (t) eiγn (t) |ψn (t)i (A.54)
with dynamic and geometric (Berry) phases

1 t
Z Z t
θn (t) = − En (t0 ) dt0 γn (t) = i hψn (t0 )|ψ̇n (t0 )i dt0 . (A.55)
h̄ 0 0

A.10.1 Landau-Zener transitions


|H12 |
If H(t) = 12 αtσz + H12 σx , we can define transition duration τ = α and Rabi
frequency ω = |Hh̄12 | . The transition is adiabatic if

|H12 |2
= ωτ  1. (A.56)
h̄|α|
The probability of diabatic passage is
|H12 |2
 
P = exp −2π . (A.57)
h̄α

A.10.2 Berry’s phase

If H depends on some time-dependent R ~ which travels through a configuration


space, Berry’s phase is
Z Rf
γn = ~
i hψn (R)|∇ ~ ~
~ |ψn (R)i · dR (A.58)
R
Ri

where the integrand is Berry’s connection A~ n (R),


~ which has gauge dependence.
~ moves in
If we integrate over a closed loop, γn becomes gauge invariant. If R
one dimension, γn = 0.

A.10.3 Molecules and Born-Oppenheimer

In molecules, energy scales are


r
m m
Eelec : Evib : Erot ∼ 1 : : (A.59)
M M
where m is the electron mass and M is the molecular mass.
The molecular Hamiltonian is
~ + Ke + VN e (R,
H = KN + VN N (R) ~ ~r) + Vee (~r). (A.60)

Assume R ~ changes adiabatically, so we can assume it’s fixed. Then solve for
electron wavefunctions
~ R (R).
(Ke + VN e + Vee )φR (~r) = Ee (R)φ ~ (A.61)

137
Tony Zhang A Useful formulae

Now assuming separability


~ ~r) = η(R)φ
ψ(R, ~ ~ (~r), (A.62)
R

we can use an effective Hamiltonian


X (pα − Aα )2
Heff = + U (R) (A.63)
α
2Mα

with
X h̄2 Z 2
X A2
α
U (R) = VN N (R) + Ee (R) + |∇R φR (r)| dr − (A.64)
α
2M α α
2M α

~ ~r)|H|ψ(R,
satisfying hψ(R, ~ ~r)i = hη(R)|H
~ eff |η(~
r)i.

A.11 Scattering
A.11.1 Partial waves and phase shifts

Given poential with finite range a, we would like to write energy eigenstates as

eikr
ψ(r) = ψincident (r) + ψscattered (r) ' eikz + fk (θ, φ) (A.65)
r
h̄2 k2
where the last part holds for r  a. We have energy E = 2m and differential
cross-section

= |fk (θ, φ)|2 . (A.66)
dΩ
Optical theorem

σ= =fk (θ = 0). (A.67)
k
Plane waves can be expanded in terms of spherical partial waves

√ X √
eikz = eikr cos θ = 4π i` 2` + 1j` (kr)Y`,0 (θ) (A.68)
`=0

where j` is a spherical Bessel function and Y`,0 is a spherical harmonic.


Define phase shifts as the δ` such that asymptotically, the total wavefunction
ψ ∼ eikz + fk (θ)eikr /r is
√ 1 1
!
4π X ` √ 1 ei(kr− 2 `π) e2iδ` e−i(kr− 2 `π)
ψ(r) = i 2` + 1Y`,0 (θ) − . (A.69)
k 2i r r
`

The scattering amplitude is related to the phase shifts:


√ ∞
4π X √
fk (θ) = 2` + 1Yl,0 (θ)eiδ` sin δ` . (A.70)
k
`=0

A consequence is that
4π X
σ= (2` + 1) sin2 δ` . (A.71)
k2
`

138
Tony Zhang A Useful formulae

If our partial waves look like

ψ` (x) = (A` j` (kr) + B` n` (kr))Y`,0 (A.72)

for r > a, then


B`
tan δ` = − . (A.73)
A`
and  
1 `π
ψ` (r) = sin kr − + δ` Y`,0 . (A.74)
kr 2
So for low-energy scattering (ka  1),

σ ' (ka)4`+2 . (A.75)

Also, using L ' h̄` ' bp = h̄bk for impact parameter b, we get b = `/k.
There shouldn’t be scattering for k` > a.
TODO TODO TODO insert hard sphere scattering

A.11.2 Integral scattering and the Born approximation

Green’s function
1 eikr
G+ (r) = − (A.76)
4π r
satisfies defining relation

(∇2 + k 2 )G+ (r − r0 ) = δ 3 (r − r0 ). (A.77)

Then we can write integral equation


Z
ψ(r) = ψ0 (r) + G(r − r0 )U (r0 )ψ(r0 ) d3 r (A.78)

Born series
Z
0
ψ(r) = eiki ·r + G+ (r − r0 )U (r0 )eiki ·r d3 r0
Z Z
+ d r G+ (r − r )U (r ) d3 r00 G+ (r0 − r00 )U (r00 )ψ(r00 ). (A.79)
3 0 0 0

To first order,
eikr
ψ(r) = eiki ·r + fk (θ, φ) (A.80)
r
where the scattering amplitude is just
Z
1 0
fk (θ, φ) = − e−iK·r U (r0 ) d3 r0 . (A.81)

For radial potentials V :
Z ∞
2m
fk (θ) = − rV (r) sin(Kr) dr (A.82)
Kh̄2 0

where K is the transfer momentum.

139
Tony Zhang A Useful formulae

A.12 Identical particles


Given single-particle state space V , define permutation operators Pα on V ⊗N

Pα |u1 i(1) ⊗ . . . ⊗ |uN i(N ) = |uα(1) i(1) ⊗ . . . ⊗ |uα(N ) i(N ) . (A.83)

These form a representation of symmetric group SN . All Pα are unitary, but not
all are Hermitian. One class of α give Hermitian operators: the transpositions.
Symmetrizer and antisymmetrizer
1 X 1 X
S= Pα A= ε α Pα (A.84)
N! α N! α

project onto SymN V and AntiN V , which are the state spaces of N -particle
ensembles of bosons and fermions, respectively.

O 1
A |ωi i(i) = det |ωi i(j) . (A.85)
i
N!

Ensembles of N identical particles are characterized by occupation numbers


of single-particle energy levels. If V has orthonormal basis |ui i, For bosons,
⊗n
O
|n1 n2 . . .iS ∝ S |ui i i . (A.86)
i

For fermions, the ni = 0, 1, by the Pauli exclusion principle.


Maxwell-Boltzmann, Fermi-Dirac, and Bose-Einstein distributions:
di 1 di
ni = E−µ ni = E−µ ni = E−µ . (A.87)
e kB T
e kB T
+1 e kB T
−1

140
Tony Zhang B Useful math

B Useful math
B.1 Trigonometric integrals
Average of sin2 and cos2 over a sphere.
Z Z
1 1 1 2
cos2 θ dΩ = sin2 θ dΩ = . (B.1)
4π 3 4π 3
To remember: sin2 θ is larger over the equatorial region, which is larger on the
sphere than the polar regions where cos2 θ is large.

B.2 Gaussian integrals


For a > 0:

Z r
2 π
e−ax dx = (B.2)
−∞ a

Z r
2 −ax2 1 π
x e dx = (B.3)
−∞ 2 a3
Z ∞ r
2 3 π
x4 e−ax dx = (B.4)
−∞ 4 a5

r
(2k − 1)!!
Z
2 π
x2k e−ax dx = (B.5)
−∞ 2k a2k+1

B.3 Gamma function


The gamma function is defined as the analytic continuation of
Z ∞
Γ(z) = tz−1 e−t dt. (B.6)
0

It generalizes the factorial on the nonnegative integers:


Γ(n) = (n − 1)! (B.7)
so that Z ∞
x
xn e− a dx = n! an+1 . (B.8)
0
The gamma function also gives the Gaussian integral with z = 21 :
  Z ∞
1 2 √
Γ = e−x dx = π (B.9)
2 −∞

B.4 Spherical harmonics

1
Y00 (θ, φ) = √ (B.10)

r
3
Y10 (θ, φ) = cos θ (B.11)

r
3
Y1,±1 (θ, φ) = ∓ sin θe±iφ (B.12)

141
Tony Zhang B Useful math

B.5 Generalized Laguerre polynomials


The general case:
n  
1 X n! α + n

n (x) = (−x)j (B.13)
n! j=0 j! n − j

For degrees n ≤ 2:


0 (x) = 1 (B.14)

1 (x) = −x + α + 1 (B.15)
1 2


2 (x) = x − 2(α + 2)x + (α + 1)(α + 2) (B.16)
2

B.6 Spherical Bessel functions


The spherical Bessel functions j` and n` are linearly independent solutions u/r
spanning the solution space of the radial equation
 2 
d `(` + 1)
ρ − + 1 u=0 (B.17)
dρ2 ρ2

where ρ = kr.
They can be defined as
 `  `
1 ∂ sin ρ 1 ∂ cos ρ
j` (ρ) = (−ρ)` n` (ρ) = −(−ρ)` . (B.18)
ρ ∂ρ ρ ρ ∂ρ ρ

To get from j` to n` , replace cos ρ 7→ sin ρ and sin ρ 7→ − cos ρ. The first two j`
and n` are
sin ρ
j0 (ρ) = (B.19)
ρ
sin ρ cos ρ
j1 (ρ) = 2 − (B.20)
ρ ρ
cos ρ
n0 (ρ) = − (B.21)
ρ
cos ρ sin ρ
n1 (ρ) = − 2 − (B.22)
ρ ρ
The n` are singular at the origin.

142

You might also like