Lidia QM2 LectureNotes
Lidia QM2 LectureNotes
2012
2
These scribbles are based on the lecture notes of Quantum Mechanics II at ETH Zurich. The idea is to have
an English version of the script, including all the steps and explanations that are missing from the original.
QM2 for dummies, if you will.
If you want to contribute, drop me an email: [email protected].
Notation
Where did that come from? Well, from many frustrated tentatives. . .
Approximation alert
[text] Comments
OS Original script
SE Schrödinger equation
DE Dierential equation
PT Perturbation theory
h.c. Hermitian conjugate
Chapter 1
Scattering theory
The goal of this section is to solve the SE for systems with spherical symmetry, i.e., with a radial potential,
V (x) = V (r). For instance, there could be a very round ion at position r = 0, or maybe a tiny sphere made out
of gelatine. It will be useful to represent this world in spherical coordinates, x = (r, θ, φ).
~2 2
− ∇ Ψ(r, θ, φ) = [E − V (r)]Ψ(r, θ, φ). (1.1)
2m
Expanding the Laplacian ∇2 in spherical coordinates, we obtain
~2
1 ∂2 1 ∂2
1 1 ∂ ∂
− r− 2 sin φ + Ψ(r, θ, φ) = [E − V (r)]Ψ(r, θ, φ). (1.2)
2m r ∂r2
r sin φ ∂φ ∂φ 2 ∂θ 2
sin
| {z }
M 2 (θ,φ)
The term M2 only depends on the angles θ and φ. Its eigenfunctions are the spherical harmonics Y`,m (which
you may have seen before in Mathematical Methods for Physics or Electrodynamics),
s
(2` + 1)(` − m)! m
Y`,m (θ, φ) = P` (cos θ) eimφ , (1.3)
4π (` + m)!
where ` and m are the quantum numbers corresponding to the angular momentum of the particle and the
z -component of that angular momentum, so they are integers such that ` ≥ 0, |m| ≤ `. P`m are the Legendre
polynomials,
(−1)m d`+m
P`m (x) = `
(1 − x2 )m/2 `+m (x2 − 1)` (1.4)
2 `! dx
(don't worry memorizing those formulas). The eigenvalue of M2 corresponding to each spherical harmonic is
given by
The spherical harmonics have some neat properties, like being all orthogonal and spanning the space of functions
on the sphere.
For now we are going to look for wave functions of the form Ψ`m = R`m (r)Y`m (θ, φ). Later on we can gener-
P
alize this: since the SE is a linear dierential equation, any linear combination of solutions, Ψ= `,m c`m Ψ`m ,
is also a solution. So far we have
~2 1 ∂ 2 1 ∂2
1 1 ∂ ∂
− r− 2 sin φ + R`m (r)Y`m (θ, φ) = [E − V (r)]R`m (r)Y`m (θ, φ) (1.6)
2m r ∂r2 r sin φ ∂φ ∂φ sin2 ∂θ2
~2 1 ∂ 2
`(` + 1)
− 2
r− R`m (r)Y`m (θ, φ) = [E − V (r)]R`m (r)Y`m (θ, φ), (1.7)
2m r ∂r r2
3
4 CHAPTER 1. SCATTERING THEORY
from which we may conclude that R(r) does not depend on m. If we x θ and φ, we can see Y`m (θ, φ) 6= 0 as a
constant and get rid of it. We will focus on nding the radial part R` (r) of the wave function,
~2 1 ∂ 2
`(` + 1)
− r − R` (r) = [E − V (r)]R` (r). (1.8)
2m r ∂r2 r2
The potential V (r) is still too general for us. To keep the hope of nding an exact solution for the equation,
we will consider only step-like potentials, ,
V0 , r ∈ [0, a0 )
V1 , r ∈ [a0 , a1 )
V (r) = . (1.9)
.
.
Vn , r ∈ [an−1 , ∞).
You could argue that that is not a very natural form for a physical potential, and you would be absoulutely
right. But it is a good rst step. Our approach will be to rst solve the SE for each region of space where the
potential is at (r ∈ [ai−1 , ai ), V (r) = Vi ). This will give us a set of solutions for that region, R`i,1 , R`i,2 , etc. We
will then combine them to nd R` (r) for the whole universe. We will do this by making sure that the global R
is still a good candidate for a wave function:
[Where do these conditions come from? Quite possibly from QM1, but I should investigate this further. To
check: Schumacher's book]
For now we will consider only the case where our particule's energy is larger than the potential, E > Vi , ∀i
(we will get rid of this assumption in Section 1.3.2). Some examples are
1 ∂2
`(` + 1) 2
r − + k R` (r) = 0 (1.12)
r ∂r2 r2
(this only works if E > Vi ; we will see the other cases in Section 1.3.2). We can simplify this equation further
if we get rid of the constant k2 . To do so, we dene a new variable,
ρ ∂2 1 ∂2
ρ := kr, r= , 2
= 2 2, (1.13)
k ∂r k ∂ρ
And now the SE is simply
k 2 ∂2 ρ
2 `(` + 1) 2
k −k + k R` (ρ) = 0 (1.14)
ρ ∂ρ2 k ρ2
1 ∂2
`(` + 1)
ρ− + 1 R` (ρ) = 0. (1.15)
ρ ∂ρ2 ρ2
Good. Now that we have simplied our problem to turn the SE into a rather standard dierential equation, we
can solve it.
1.3. INTERLUDE: RADIAL POTENTIAL 5
This proof is fastidious, full of miracles ( ), and not particularly enlightening. Feel free to skip it in order
to preserve your soul and enthusiasm for quantum mechanics.
Case `=0
When ` = 0, the DE is just
1 ∂2
ρR` (ρ) + R` (ρ) = 0. (1.16)
ρ ∂ρ2
This equation could not get more textbook than this, so we can look it up and nd that it has two solutions:
sin ρ − cos ρ
R0 (ρ) = , R0 (ρ) = , (1.17)
ρ ρ
where the sign of each solution is just a convention. At rst sight, we can see that the second solution diverges
for ρ → 0, so when later we choose the physical solutions of the SE, we will discard that one for small ρ. Later.
Case `>0
In this case, our DE is
1 ∂2
`(` + 1)
ρR ` (ρ) + − + 1 R` (ρ) = 0. (1.18)
ρ ∂ρ2 ρ2
We will start with a substitution, dening a new function χ` (ρ) such that
We will omit the (ρ)" for simplicity's sake. The second derivative of Eq. 1.18 becomes
∂2 ∂ 2 `+1
ρR` (ρ) = ρ χ`
∂ρ2 ∂ρ2
= `(` + 1)ρ`−1 χ` + 2(` + 1)ρ` χ0` + ρ`+1 χ00` .
`(` + 1)
0 = ρ `(` + 1)ρ`−1 χ` + 2(` + 1)ρ` χ0` + ρ`+1 χ00` + − ρ` χ`
+ 1
ρ2
= `(` + 1)ρ`−2 χ` + 2(` + 1)ρ`−1 χ0` + ρ` χ00` −`(` + 1)ρ`−2 χ` + ρ` χ`
1 2(` + 1) 0
multiply by ,→ 0 = χ` + χ` + χ00` . (1.20)
ρ ρ
This is a decent equation, but not yet the recursive solution we are looking for. Hold on.
∂ 2(` + 1) 0 2(` + 1) 00
take ,→ 0 = χ0` − χ` + χ` + χ000
`
∂ρ ρ2 ρ
2(` + 1) 2(` + 1)
substitution: χ0` = χρ ,→ 0 = χρ − χρ + [χρ]0 + [χρ]00
ρ2 ρ
2(` + 1) 2(` + 1)
= χρ− χ+ χ+2(` + 1)χ0 + χ00 ρ+2χ0 + 0
ρ ρ
= χρ+2(` + 2)χ0 + χ00 ρ
1 2(` + 2) 0
multiply by ,→ 0 = χ + χ + χ00 .
ρ ρ
6 CHAPTER 1. SCATTERING THEORY
χ0`
χ0`+1 = χ = , (1.22)
ρ
which is precisely the kind of recursive solution we were looking for. Going back all the way ` → `−1 → · · · → 0,
we get
` `
−−−−−−`−→
1 ∂ 1 ∂
χ` = χ0 R` = ρ χ` R` (ρ) = ρ` R0 (ρ). (1.23)
ρ ∂ρ ρ ∂ρ
Hooray!
Remember how we had two solutions for R0 ? This gives origin to two families of solutions: the Bessel functions,
j` (ρ), and the Neumann functions, n` (ρ). Here follow their expression and behaviour for small and large ρ.1
Again, the minus signs are just a practical convention.
sin ρ
`=0 j0 (ρ) = ρ n0 (ρ) = − cosρ ρ
` `
1 ∂ sin ρ 1 ∂ cos ρ
general j` (ρ) = (−1)` ρ` ρ ∂ρ ρ n` (ρ) = (−1)`+1 ρ` ρ ∂ρ ρ
ρ` (2`−1)!!
ρ→0 j` (ρ) ≈ (2`+1)!! n` (ρ) ≈ ρ`+1
1 π`
n` (ρ) ≈ − ρ1 cos ρ − π`
ρ→∞ j` (ρ) ≈ ρ sin ρ − 2 2
Hankel functions
We found two families of solutions for our DE. Remember that any linear combination of solutions of a linear
dierential equations is a solution (it might not be a physical solution, but we will get to that in a moment).
Well, there are two linear combinations of Bessel and Neumann functions that will be especially useful for
scattering problems. So useful indeed, they get their own name: the Hankel functions.
(1) (2)
h` (or h` ) h∗` (or h` )
denition h` (ρ) = j` (ρ) + in` (ρ) h∗` (ρ) = j` (ρ) − in` (ρ)
` `
eiρ e−iρ
general h` (ρ) = −i (−1)` ρ` 1 ∂
ρ ∂ρ ρ h∗` (ρ) = i (−1)` ρ` 1 ∂
ρ ∂ρ ρ
Now we are nally ready to look at some concrete, if contrived, physical systems.
1 For a proof of the asymptotic behaviour, see Exercise 2 of the 1st series.
1.3. INTERLUDE: RADIAL POTENTIAL 7
1 ∂2
`(` + 1) 2m
2
r− 2
R`in (r) = − 2 [E + V0 ]R`in (r), E + V0 > 0. (1.24)
r ∂r r ~
We can use the same substitution as before,
p
2m(E + V0 )
q= , (1.25)
~
and get the two families of solutions, j` (qr) and n` (qr). This region includes the point r = 0, so we cannot have
solutions that diverge for qr → 0: we can discard all the Neumann functions, which means that the physical
solution inside the well is simply a multiple of j` :
R`in (r) = A` j` (qr). (1.26)
1 ∂2
`(` + 1) 2m
r− R`out (r) = − 2 ER`out (r), E < 0. (1.27)
r ∂r2 r2 ~
√
Now we have a small problem. We cannot dene k = 2mE/~ if we want k to be real. Let us use instead
p
2m(−E)
k= , (1.28)
~
which turns the SE into
1 ∂2
`(` + 1)
r − − k 2
R`out (r) = 0. (1.29)
r ∂r2 r2
Now, this is slightly dierent from our familiar Eq. 1.12. If we want to end up with something that looks like
Eq. 1.15, we need to dene the variable ρ to be
1 ∂2 −1 ∂ 2
ρ := ikr, r= ρ, 2
= 2 , (1.30)
ik ∂r k ∂ρ2
and then we have
ik 2 ∂ 2 ρ
2 `(` + 1)
−k −k −k R`out (ρ) = 0
2
(1.31)
ρ ∂ρ2 ik ρ2
1 ∂2
`(` + 1)
ρ− + 1 R`out (ρ) = 0. (1.32)
ρ ∂ρ2 ρ2
8 CHAPTER 1. SCATTERING THEORY
This is precisely the same equation as before, and therefore has the same (mathematical) solutions, j` (ikr) and
n` (ikr). To pick the physical solution, let us look at the asymptotic behaviour of these functions when r → ∞.
1 π`
j` (ikr) ≈ sin ikr −
ikr 2
− i ix −i −i
ei(ikr− 2 ) − e−i(ikr− 2 )
π` π`
e − e−ix ,→
sin x = =
2 kr 2
!
−1 −kr −i π` kr i π`
= e e 2 − |{z} e e 2 ,
2kr |{z}
→0 →∞
Fortunately for us, not all solutions diverge. We just have to look for a dierent basis in which to expand our
general solution. For instance, we can try the Hankel functions,
i i(ikr− π`
2 ) =
−1 −kr i π`
h` (ikr) ≈ − e e e 2, (1.33)
ikr kr |{z}
→0
i −i(ikr− π`
2 ) =
1 π`
h∗` (ikr) ≈ e ekr e−i 2 . (1.34)
ikr kr |{z}
→∞
This is great: h∗` (ikr) diverges, h` (ikr) does not, so the physical solution outside the well is just a multiple of
the latter,
[Auf der Mauer, auf der Lauer sitzt 'ne kleine Welle.]
Now we can look for the coecients A` and B` . Since V0 < 0, we know three things:
∂ in ∂ out
R (r) = R (r) ⇔
∂r ` r=a ∂r ` r=a
⇔ A` q j`0 (qa) = ikB` h0` (ika). (1.37)
P
3. The wave function is normalized. This applies to the total wave function, Ψ= `m c`m R` (a)Y`m (θ, φ).
Depending on the coecients {c`m }, we may x A` , and get B` from the rst condition.
A` j` (qa) = B` h` (ika)
A` h` (ika)
=
B` j` (qa)
` i(ikr)
1 ∂ e
−i (−1)` (ikr)` ikr ∂(ikr) ikr
= ` .
1 ∂ sin(qr)
(−1)` (qr)` qr ∂(qr) qr
r=a
1.3. INTERLUDE: RADIAL POTENTIAL 9
Now, this might be doable, but it is more than a little ugly, and overall not that interesting. Anyway, as an
example, let us solve it for ` = 0,
i(ikr)
A0 −i e ikr
= sin(qr)
B0
qr r=a
q e−ka
=− . (1.38)
k sin(qa)
Now A0 will depend on the coecients c`m . But there is something more interesting we can do with these
continuity conditions.
[it would be cute to have a drawing of |R0 (r)|]
We saw that the wave function vanishes quickly outside the well (it goes approximately with e−kr /r). This is
because the energy of the particle is smaller than the potential outside the well (E < V = 0). The only place
where the wave function is stable is inside the well, which is why we call it a bound state : the wave is bound to
the well. It turns out that if the well is not deep enough, there will be no wave function anywhere: we will nd
a condition for the minimum depth the well such that if V0 < Vmin ⇒ A0 = B0 = 0 ⇒ R0 (r) = 0∀r. This can
be generalized for arbitrary `, but you will get the idea from here.
First a few mathematical tricks. The factor 1/ρ in the expressions of R0in (r) and R0out (r) makes the derivatives
1
a little annoying. It would be lovely to get rid of it, and here is how we do it. We want R0 (r) ∈ C
1
(i.e. both R0 and its rst derivative are continuous). In general, if we have two functions f (r), g(r) ∈ C , then
f (r) g(r) ∈ C 1 . Well, take f (r) = r, g(r) = R0 (r), and the condition for continuity of R0 implies
(
1 A0 sin(qr), r < a
C 3 r R0 (r) = , (1.39)
B0 e−kr , r>a
√
a 2m
We can simplify this with two subsitutions,
√ C = ~ >0 and
x = C E + V0 > 0, to get
√
x tan x = − C −E (1.45)
Let us nd
| {z }
D>0
√
a 2m p π
E + V0 > (1.47)
~ 2
π 2 ~2 π 2 ~2
V0 > 2 − E = 2 +|E|. (E < 0) (1.48)
8a m |8a{zm}
Vmin
In other words, if V0 < Vmin + |E|, then we cannot have tan(qa) = −q/k , and so, due to Eq. 1.41 we need
A0 = B0 = 0, which means that R0 (r) = 0, ∀r.
(
−V0 , r < a
V (r) =
0, r ≥ a,
As before,
p
in
2m(E + V0 )
R` (r) = A` j` (qr), q= (1.49)
~
r→0
Note that, again, n` (qr) −−−→ −∞ and thus n` cannot appear as part of the wave function.
√
2mE
R` (r) = B` j` (kr) + C` n` (kr), k= (1.50)
~
Since in this case r→0 is not included in the domain, n` can be part of the wave function.
C`
Our goal in this analysis is to nd the constants A` , B` and C` . Let us start by looking at the ratio
B` .
B` `π c `π
R` (r) ≈ sin kr − − cos kr − (1.51)
kr 2 kr 2
B `π C `π
= sin kr − − cos kr − (1.52)
kr 2 B 2
We dene
C
tan δ` = (1.53)
B
With this denition, we have
B `π `π
R` (r) = sin kr − + tan δ` cos kr − (1.54)
kr 2 2
Using
we nally get
B 1 `π
R` (r) = sin kr − + δ` . (1.56)
kr cos δ` 2
Remember that this is only true for large r. We can nd for instance δ0 ,
k
δ0 = tan−1 tan (qa) − ka. (1.57)
q
[For further physical interpretation, check the original script and suggested books (I was getting a bit tired
of this kind of problems when I got here).]
(
∞, r ≤ a
V (r) =
0, r > a.
The wave function Ψ is thus 0 inside the sphere. On the outside, we can expand R` in the basis of the Hankel
(1) (2)
functions h` := h` and h∗` := h` :
√
smE
R` (r) = Ah∗` (kr) + Bh` (kr), k= (1.58)
~
B
= A h∗` (kr) + h` (kr) . (1.59)
A
B
For now we are only interested in the fraction
A between the two components. Up to a global phase and
normalization (which will come later), we can write
Now we can reuse the notion of probability ux j, introduced in the beginning of the chapter (see original
script). Since no particles are created or destroyed in the scattering process, the radial component jr (r) = 0∀r
. We have
~ ∂ ∂
0 = jr (r) = R`∗ (r) R` (r) − R` (r) R`∗ (r)
2im ∂r ∂r
~
(h` + S`∗ h∗` ) h∗` 0 + S` h0` − (h∗` + S`∗ h` ) h0` + S` h∗` 0
=
8im
~
h` h∗` 0 + h` S` h0` + S`∗ h∗` h∗` 0 + |S` |2 h∗` h0` − h∗` h0` − h∗` S`∗ h∗` 0 − S` h` h0` − |S` |2 h` h∗` 0
=
8im
~
h` h∗` 0 − h∗` h0` + |S` |2 h∗` h0` − |S` |2 h` h∗` 0
=
8im
~
h` h∗` 0 − h∗` h0` 1 − |S` |2 .
=
8im
Since this has to hold for all r and h` is r-dependent, we can conclude that
|S` |2 = 1, (1.61)
S` = e2iδ` . (1.62)
i i
h` (ρ) = − ei(ρ− 2 ) h∗` (ρ) = − e−i(ρ− 2 ) ,
`π `π
(1.63)
ρ ρ
we get
and thus
h∗` (ka)
e2iδ` = − (1.68)
h` (ka)
j` (ka) − in` (ka)
=− . (1.69)
j` (ka) + in` (ka)
We can express this relation using a the tangent [this was in one of the exercise series],
j` (ka)
tan δ` = (1.70)
n` (ka)
(ka)`
j` (ka) ≈ (1.71)
(2` + 1)!!
(2` − 1)!!
n` (ka) ≈ (1.72)
(ka)`+1
which leads to
(ka)2`+1
tan δ` ≈ . (1.73)
(2` + 1)[(2` − 1)!!]2
For ` ≥ 1, this decays rapidly. When computing the cross section σ, we have to sum over all `. In a rst
approximation, this allows us to only consider ` = 0. [More about this in the exercise series.]
∞ X
X `
eikx = c`m (k)j` (kr)Y`m (θ, φ) (1.74)
`=0 m=−`
Since we are considering the rotationally symmetric case with kx = kx cos θ, φ cannot be part of the expression.
Thus only the m=0 terms will remain. We use the simplied notation
1 d` 2
P` (x) = (x − 1)` (1.77)
2` `! dx`
and satise the orthogonality relation
Z 1
2δ``0
dxP` (x)P`0 (x) = (1.78)
−1 2` + 1
∞ r
ikr cos θ
X 2`0 + 1
e = A`0 j`0 (kr)P`0 (cos θ) (1.79)
4π
`0 =0
To nd A` j` (kr), we project eikr cos θ onto P` (cos θ), which basically means that we multiply both sides of
the expression by P` (cos θ) and integrate over d cos θ. Using the orthogonality relation (1.78) of the Legendre
polynomials, we thus get (setting z := cos θ in the integral)
1
p Z
4π(2` + 1)
A` j` (kr) = dzP` (z)eikrz . (1.80)
2 −1
X (ikrz)`
eikrz = (1.81)
`!
`
Looking at the leading order term in the Legendre Polynomial from (1.77)
(2`)! `
z + O z `−1
P` (z) = ` 2
(1.82)
2 (`!)
which, after solving for z` and plugging it back into the expansion for eikrz we get
!
1
p Z
4π(2` + 1) X (2`)!
(ikr)` − O (kr)`−1
A` j` (kr) = dzP` (z) (1.83)
2 −1 2` `!
`0
Using again the orthogonality relation of the Legendre polynomials (1.78), we then nd
p
4π(2` + 1) 2`+1 `!
(ikr)` + O (kr)`−1
A` j` (kr) = (1.84)
2 (2` + 1)!
2` `!
j` (kr) ≈ (kr)` (1.85)
(2` + 1)!
we nally nd
p
A` = i` 4π(2` + 1) (1.86)
∞
X
eikr cos θ = i` (2` + 1)j` (kr)P` (cos θ). (1.87)
`=0
Chapter 2
Our setting is simply a particle in an EM eld (light). We want to nd out how light is scattered by the paricle.
If you shoot a light beam in the direction of the particle, will it change direction? Colour? Will it excite the
particle? What is the probability of each process? So far (in previous lectures and original script) we looked at
one particular scattering process. We will now look at few more possibilities. First, recap.
The Hilbert space of the particle is Hp the one of the EM eld is HEM . The global Hilbert space is, naturally,
H = Hp ⊗ HEM . (2.1)
We quantized the EM eld. Fist, we drew an imaginary cubic box of side L (Fig. 2.1). Then we imposed
boundary conditions: we said that wave functions had to be zero at the borders of the box. We can do this by
making our box larger than the universe (and assuming that nothing exists outside the universe).
Figure 2.1: Giant box in real space, and some of the possible waves that full the boundary
conditions from Eq. 2.2.
k k k k
ψEM (0, 0, 0) = ψEM (L, y, z) = ψEM (x, L, z) = ψEM (x, y, L) = 0. (2.2)
This implies that the wave function of the EM eld is a linear combination of plane waves with k vectors of the
form
n
2π x
k= ny , n i ∈ Z. (2.3)
L
nz
2π
The allowed states in momentum space form an innite lattice of spacing
L . Photon's polarization, λ, is
independent of momentum. We can write the Hilbert space of the EM eld as a tensor product of the spaces of
N
all the allowed momenta and polarizarion, HEM = k,λ Hk,λ .
To describe a particular EM eld (i.e., a particular state of this Hilbert space), we populate the allowed sites
in momentum space with actual photons. We denote the vacuum (a state where there are no photons) by |0i,
and then act on this state with creation operators. For instance, â†kλ |0i is a state with exactly one photon with
momentum k and polarization λ. Here is another example: we have an initial eld |ψi with lots of photons.
Now it suers a proccess that should change the momentum of a photon with polarization λ from k to k0 . The
†
new state is âk0 λ âkλ |ψi. First we have an anihilation operator that destroys the photon with momentum k, and
0
we create a photon with momentum k .
14
2.6. LIGHT SCATTERING RELOADED 15
The anihilation operator âkλ acts only on the Hilbert space Hk,λ ; explicitly, we have
and are otherwise characterized early in the original script. [If you are not comfortable with these operators,
go back to where they are introduced and solve some of the earlier exercise series until you feel you understand
them well.] Note that this way of organizing the Hilbert space of the EM eld is very dierent from the way
we describe the particle's space. There we act on the Hilbert space with operators like r̂ and p̂, whereas here
everything is done in terms of ladder operators. For instance, to measure the average energy of the eld, we
dene the operator
Ek â†k,λ âk,λ ,
X
Ê = hEiψ = hψ|Ê|ψi, Ek = ~k. (2.6)
kλ
This operator simply counts the number of photons in each mode (a pair (k, λ)), and multiplies it by the mode's
energy. Compare this with the usual way of computing the average energy of a free particle in state |φi,
p̂2
hEiφ = hφ| |φi.
2m
There we have to act with the operator p̂ (eectively a gradient) on the wave-function, which is usually harder
than counting photons in a given mode.
We denote the dierent states of the particle by |0i, |1i, . . . , |ni, etc., in increasing order of energy. We denote
the states of the EM eld by the number of photons of each mode, like this: |Nk1 ,λ1 , . . . , Nk,λ , . . .i. Sometimes
we are only interested in one or two modes, for instance, if we are looking for processes that change the number
of photons in those modes (like photon absorption or deection). In that case we may label the state only by the
relevant modes: for instance, if a process takes one photon from mode (k, λ) to mode (k0 , λ0 ), we can represent
this as |Nk,λ , Nk0 ,λ0 i −→ |Nk,λ − 1, Nk0 ,λ0 + 1i. All the other modes are implicitly left unchanged.
Later on (Chapter 4) we will express the Hilbert space of particles in this new formalism too (second
quantization). For now, here is a table with some characteristics of the two spaces.
Particle EM eld
N
Hilbert space Hp HEM = k,λ Hk,λ
Operators r̂, p̂ âk,λ , â†k,λ ,
N̂k,λ = â†k,λ âk,λ
Typical states |ni |Nk1 ,λ1 , . . . , Nk,λ , . . .i
(labelled by energy) (labelled by # photons/mode)
Global Hamiltonian
The total Hamiltonian was derived earlier in the original script. It can be written as
p̂2
Ĥp := + V̂0 (r̂) , (2.8)
2m | {z }
e.g. atom
|{z}
free particle
(see for instance Series 5, exercise 1). The interaction term that couples the particle and the radiation is again
given by three terms,
Goal
In general we are interested in scattering processes that excite the particle and change the EM eld in some
way,
|0ip ⊗ |Nk1 ,λ1 , . . . , Nk,λ , . . .iEM =⇒ |nip ⊗ |Nk0 1 ,λ1 , . . . , Nk,λ
0
, . . .iEM . (2.13)
| {z } | {z }
=:|ii =:|f i
We want to know which of these transitions processes are allowed, and their probability of hapenning. Ideally
we would solve the SE for the global system, but that is too hard with this complex Hamiltonian. We can
however nd the most likely transitions if we apply perturbation theory to each of the interaction terms of the
Hamiltonian. Our plan is the following:
Why don't we apply second order to all the terms? As far as I can see, it is because it would be a long,
tedious calculation that would not bring us any new transitions or insights.
Perturbation theory to rst order results in Fermi's Golden Rule. There is a summary of the derivation of
the rule in the QM1 OS, pages 127130. Short summary: we write the SE as a function of the interaction
Hamiltonian, Hint , and Taylor-expand it on time. Then we consider only the rst-order terms and compute the
probability of a transition from state |ii to state |f i. This is called Fermi's Golden Rule, and the probability is
given by
2π 2
Γi→f = |hf |Hint |ii| ρ(Ef ), (2.14)
~
where ρ(Ef ) is the density of states of the nal state (usually something like δ(Ef − Ei )). We will explain how
perturbation theory works in detail in Section 2.6.2, when we introduce the second-order terms.
We quantized the of the interaction Hamiltonian in previous lectures (Eq. 2.4.17 of the OS),
r
e X 2π~
Ĥint,1 = −√ ĵ(−k) · e(k, λ) ⊗ âk,λ + ĵ(k) · e(k, λ)∗ ⊗ â†k,λ (2.15)
L3 k,λ ωk
We applied Fermi's Golden Rule and found the transition probability Γi→f for the transition
What is happening in this transition? The particle is absorbing one photon in mode (k, λ) and using its energy
to go to the excited state |ni (see Fig. 2.2a). This process can be represented as a Feynman diagram (introduced
in Fig. 2.2b).
2.6. LIGHT SCATTERING RELOADED 17
x
γ : k, λ
e− : |0i |ni
t
k, λ |0i |ni (b) Photon absorption described as a Feynman diagram.
Time ows from left to right. The vertical axis dierentiates
(a) Left (before the interaction) , we have a pho- between the two Hilbert spaces: EM eld on top, particle
ton about to hit the photon hits the particle. on bottom. The wobbly line represents the photon, and the
On the right, the photon was absorbed by the straight line the particle (an electron). [Note: sometimes
particle, which is now in an excited state. Feynman diagrams are rotated by 90°(e.g., in the script).]
Figure 2.2: A particle absorbs a single photon in mode (k, λ), going from state |0i to |ni. Here are
two dierent pictorial representations of the transition.
Now we are going to write the second interaction term of the Hamiltonian in our new formalism. The original
form of this term is
Z Z
e 3 2 e
Ĥint,2 = d r ρ̂(r̂) Â (r̂) = d3 r ρ̂(r̂) ⊗ Â2 (r̂),
2mc2 2mc2
because the operator Â(r) = 1̂p ⊗ Â(r)EM acts only on HEM , and ρ̂(r) = ρ̂(r)p ⊗ 1̂EM acts only on Hp .
Remember that the vector potential  is dened as
s
1 X 2π~c2
Â(r)EM := √ âk,λ e(k, λ)eik·r + â†k,λ e∗ (k, λ)e−ik·r . (2.17)
L3 ωk
k,λ
We saw earlier in the OS that this denition satises B̂ = ∇ ∧ Â. Squaring the vector potential, we obtain
2
2π~c X 1 0
Â2 (r) = √ âk,λ e(k, λ)eik·r + h.c. âk0 ,λ0 e(k0 , λ0 )eik ·r + h.c. . (2.18)
L3 ωk ωk0
k,λ,k0 ,λ0
e 2π~c2
Z X 1 0
Ĥint,2 = d3 rρ̂p (r̂) ⊗ √ âk,λ e(k, λ)eik·r + h.c. âk0 ,λ0 e(k0 , λ0 )eik ·r + h.c. .
2mc2 L3 ωk ωk0
k,λ,k0 ,λ0
(2.19)
We are ready to apply Fermi's Golden Rule. But what initial and nal states should we consider? Right
now it is not obvious, but the calculations ahead will show us that this interaction term leads to a transtion
where the particle absorbs a photon, goes to an excited state, and emits a dierent photon (see Figs. 2.3a
and 2.3b),
|0ip ⊗ | . . . , Nk,λ , . . . , Nk0 ,λ0 = 0, . . .iEM =⇒ |nip ⊗ | . . . , Nk,λ − 1, . . . , Nk0 ,λ0 = 1, . . .iEM . (2.20)
Initial state: particle in ground state, no photons in mode (k0 , λ0 ) [for simplicity], Nk,λ photons in mode (k, λ)
and whatever in other modes. Final state: particle in excited state |ni, one photon in mode (k0 , λ0 ) and one
less photon in mode (k, λ). The other modes are not aected. In short notation,
0 i,
|ii = |0ip ⊗ | Nk,λ , |{z} |f i = |nip ⊗ | Nk,λ − 1, |{z}
1 i. (2.21)
| {z } 0 0 | {z } 0 0
(k,λ) (k ,λ ) (k,λ) (k ,λ )
We will see for what values of n, k and k0 this transition is allowed, and with what probability.
The matrix element from Fermi's Golden Rule (Eq. 2.14) comes
Z
e
hf |Ĥint,2 |ii = hn|hNk,λ − 1, 1| d3 r ρ̂(r̂) ⊗ Â 2
(r̂) |0i|Nk,λ , 0i
2mc2
Z
e
= d3 rhn|ρ̂(r̂)|0i hNk,λ − 1, 1|Â2 (r̂)|Nk,λ , 0i. (2.22)
2mc2
18 CHAPTER 2. INTERACTION BETWEEN LIGHT AND MATTER
x
k, λ k0 , λ0
k0 , λ0
Figure 2.3: A particle, originally in state |0i, absorbs a photon with momentum k. It goes to
excited state |ni and emits a photon with energy εk0 = E(n) − E(0) − εk .
You can check that only two terms of that sum are non-vanishing: those with operators âk,λ â†k0 ,λ0 and â†k0 ,λ0 âk,λ
0 0
(in other words, operators that create a photon in mode (k , λ ) and destroy one in mode (k, λ)). In the end,
we are left with
1 2π~c2 0
hNk,λ − 1, 1|Â2 (r̂)|Nk,λ , 0i = 2e(k, λ) · e∗ (k0 , λ0 )ei(k−k )r Nk,λ .
p
3
√
L ωk ωk0
2π~c2
Z
i(k−k0 )r
Nk,λ e(k, λ) · e∗ (k0 , λ0 ) hn| 3
p
hf |Ĥint,2 |ii = r0 3 √ d r ρ̂(r) e |0i
|{z} L ωk ωk0
e2
mc2
2π~c2
Nk,λ e(k, λ) · e∗ (k0 , λ0 ) hn|ρ̂k0 −k |0i.
p
= r0 3
√ (2.23)
L ωk ωk0
Here, r0 is the constant historically known as the electron radius, and ρ̂k0 −k is the Fourier transform of ρ̂(r).
Finally, we use the Golden Rule (Eq. 2.14) to obtain the transition probability,
2π 2
Γ0→n,(k,λ)→(k0 ,λ0 ) = |hf |Hint,2 |ii| ρ(εf )
~
2
2π~c2 r0
2π 2 2
= √ Nk,λ |e(k, λ) · e∗ (k0 , λ0 )| |hn|ρ̂k0 −k |0i| δ(En + ~ωk0 − (E0 + ~ωk )).
~ L3 ωk ωk0
(2.24)
dσ 1 X X
= Γ0→n,(k,λ)→(k0 ,λ0 ) . (2.25)
dΩk0 jinc n 0
k ∈dΩk0
Here, jinc is the ux of incoming photons: the photons that we send in mode (k, λ) to excite the particle. We
have Nk,λ photons moving at the speed of light in a box of side L, so the ux is jinc = cN k
L3 . We sum over all
the photons with momentum compatible with the small section dΩk0 (photons going in the same direction but
0
with dierent |k|). Assume that the detectors know which λ hit them.
2.6. LIGHT SCATTERING RELOADED 19
k0 , λ0
k, λ |0i |ni
Figure 2.4: We place many small detectors in a circle around the particle. The photon emitted
with momentum k0 will hit the detector at section dΩk0 . We repeat the experiment many times,
and nd the statistics of photons detected for each angle.
dσ X L3 X
= Γ0→n,(k,λ)→(k0 ,λ0 )
dΩk0 cNk
n k0 ∈dΩk0
X 8π 3 ~r2 c3 X 1 2 2
= 0
|e(k, λ) · e∗ (k0 , λ0 )| |hn|ρ̂k0 −k |0i| δ(En + ~ωk0 − (E0 + ~ωk ))
L3 ωk ωk0
n k0 ∈dΩk0
We need to compute the sum over all k0 ∈ dΩk0 . We will approximate the sum (in general discrete
values) by an integral [Riemann sum from calculus]
3 Z
X L
f (k0 ) ≈ d3 k0 f (k0 ). (2.26)
2π
k0
Remember that
2π
k0 = 0
L (nx , ny , nz ). For large L, neighbouring k s are very close together, so this approximation
is reasonable. Now we can transform the Cartesian coordinates to spherical coordinates, easier to relate to our
section,
2 ωk2 0
d3 k0 = |k0 | dΩk0 d |k0 | = dΩk0 dωk0 ,
c3
so we have
∞
L3
X Z
f (k0 ) ≈ dωk0 ωk2 0 f (k0 ). (2.27)
8π 3 c3 0
k0 ∈dΩk0
X 8π 3 ~r2 c3 L3 Z ∞
dσ 1 2 2
= 0
dωk0 ωk2 0 |e(k, λ) · e∗ (k0 , λ0 )| |hn|ρ̂k0 −k |0i| δ (En + ~ωk0 − E0 − ~ωk )
dΩk0 n
L3 ωk 8π 3 c3 0 ω k0 | {z }
En −E0
1
~ δ (ωk0 + ~ −ωk )
X ~r2 Z ∞
2 2 1 E n − E 0
= 0
dωk0 ωk0 |e(k, λ) · e∗ (k0 , λ0 )| |hn|ρ̂k0 −k |0i| δ ωk0 + − ωk
n
ωk 0 ~ ~
| {z }
delta function
X ωk0 2 2
= r02 |e(k, λ) · e∗ (k0 , λ0 )| |hn|ρ̂k0 −k |0i| δ~ωk0 +En ,E0 +~ωk .
n
ωk | {z }
Kronecker delta
2
We are almost done. All that is left to calculate is the term |hn|ρ̂k0 −k |0i| . This depends on the properties
of the particle, namely on states {|nip }. We will compute it for a specic case.
Example Suppose that the particle is a free electron, such that each state is described by a plane wave,
|ni = √1 eiqn r . We know that ρ̂(r) = δ(r − r̂). The Fourier transform of this delta function is
L 3
Z
0 0
ρ̂k0 −k = d3 rδ(r − r̂)e−i(k −k)·r = e−i(k −k)·r̂ . (2.28)
L3
20 CHAPTER 2. INTERACTION BETWEEN LIGHT AND MATTER
What follows is a general description of perturbation theory to second order in time. Later we will apply this
to our concrete problem. A good reference (in German) is Manfred Sigrist's QM2 lecture notes, pages 117124.
We start from a time-dependent Hamiltonian
Here, Ĥ0 is the initial time-independent Hamiltonian (imagine something simple, like a free particle or a hydro-
gen atom). V̂ (t) is a small perturbation, compared with Ĥ0 (e.g. the EM-eld), which is "swiched on" at time
t = t0 and o again at time t = τ,
V̂ (t) = Θ(t0 )[1 − Θ(τ )] V̂ .
Assume that we know the stationary solution of the initial Hamiltonian,
Summary of assumptions: (τ − t0 ) small, ||V̂ || ||Ĥ0 ||, solution for Ĥ0 known, initial state |m(t)i.
where |ψ(t)i is the actual state of the system (the solution of the SE). In particular, just before we turn on V̂
(at t = t0 ), the interaction state is
iĤ0 t0
|ψ̃(t0 )i = e ~ |m(t0 )i = |mi. (2.35)
where T̂t is the time-ordering operator. Now we expand the exponential, obtaining
t0
"Z # !
Z t t Z
1 1
|ψ̃(t)i = 1̂ + 0 0
dt V̂i (t ) + dt 0 00 0
dt V̂i (t )V̂i (t ) + O(V̂ ) |mi. 00 3
(2.38)
i~ t0 (i~)2 t0 t0
P
This means that the nal solution is a linear combination of terms |ψ̃(t)i = j |ψ̃j (t)i, where |ψ̃j (t)i is the
solution for the j th term of the expansion. Remember that we are looking for the probability of transition from
state |mi to state |ni:
2
Prob(|mi → |ni) = |hn(t)|ψ(t)i|
2
= hn|ψ̃(t)i
2
X
= hn| |ψ̃j (t)i
j
2
X
= hn|ψ̃j (t)i . (2.39)
j
Rt R t0
the second-order term, |ψ̃2 (t)i = 1
(i~)2 t0
dt0 t0
dt00 V̂i (t0 )V̂i (t00 ) |mi, will be examined here;
A quick note about taking only the second-order term. When we apply this to our problem of light-matter
interaction, we will see that the contribution from the 0th- and rst-order terms is zero for the relevant transition
and interaction potential. In general,
(
(2) = Prob(|mi → |ni)
2 if hn|ψ̃j (t)i = 0, ∀j 6= 2
Prob (|mi → |ni) := hn|ψ̃2 (t)i (2.40)
| {z } ≈ Prob(|mi → |ni) if hn|ψ̃0 (t)i = hn|ψ̃1 (t)i = 0 and ||V̂ || small.
contribution from 2nd term
since |ni, |mi are time-independent. We will use two tricks to solve this integral .
Sort of a dirty trick. We assume V̂ (t > t0 ) = V̂ eηt , where η = limx→0 x =: 0+ . This will make our
integrals converge and it does not really change V̂ by much, because η and t are very small (remember that we
turn o V̂ quickly); in addition V̂ should be time-independent. It follows that
Quite a clever, elegant trick. In Eq. 2.41, the term hn|V̂i (t0 )V̂i (t00 )|mi is hard to compute directly (in
2
general we may not know how to write V̂i ). So we insert an identity operator between the two V̂i 's. We can
1̂ = 0
P
write the identity as ` |`ih`| for any orthonormal basis (h`|`ni = δ`,` ) of our Hilbertospace. We will choose
0
this basis to be the eigenvalues of Ĥ0 (our stationary solutions), |`i ∈ H : Ĥ0 |`i = ε` |`i . The matrix element
becomes
!
hn|V̂i (t )1V̂i (t )|mi = hn|V̂i (t )
X
0 00 0
|`ih`| V̂i (t00 )|mi
`
X
= hn|V̂i (t )|`i h`|V̂i (t00 )|mi,
0
(2.44)
`
which is easier to calculate, because now we only have matrix elements of a single operator.
Using both tricks, and setting t0 = 0 for simplicity, we can compute the total matrix element in Eq. 2.41,
Z t Z t0
1 X 0 i 0 0 00 i 00 00
hn|ψ̃2 (t)i = hn|V̂ |`ih`|V̂ |mi dt exp (ε n − ε ` )t + ηt dt exp (ε ` − ε m )t + ηt
(i~)2 t0 ~ t0 ~
`
So, the probability of transition at time t (at least the contribution from the second-order term) is
2
i 2 e4ηt X hn|V̂ |`ih`|V̂ |mi
Prob
(2)
(|mi → |ni) = e ~ (εm −εn )t 2 . (2.45)
|εm − εn + 2i~η| εm − ε` + i~η
`
| {z }
=:A
d
Γm→n = Prob(|mi → |ni)
dt
d e4ηt
= A
dt |εm − εn + 2i~η|2
4ηe4ηt
= 2A
|εm − εn + 2i~η|
4ηe4ηt
≈A since η → 0
(εm − εn )2 + (2i~η)2
A η
= 2 lim εm −εn 2
~ η→0 ( 2~ ) + η 2
+
A εm − εn
= 2δ
~ 2~
2A
= δ (εm − εn )
~
2 X hn|V̂ |`ih`|V̂ |mi
= δ (εm − εn ) . (2.46)
~ εm − ε` + i~η
`
We are nished. What can we conclude? What do those terms with |`i mean? Here are a couple of ways to
think about it:
1. Our system goes from |mi to |ni through intermediate states |`i.
The |`i's that more likely occur are the ones with energy close to that of |mi due to the term
1
, (2.47)
εm − ε` + i~η
but energy violations are possible from |mi to |`i. You might hear in particle physics something along the
lines of:
but that's ok, because of the energy-time uncertainty relation and we can never detect |`i in
the laboratory anyway. . .
2.6. LIGHT SCATTERING RELOADED 23
This interpretation makes it easy to visualise the transition going through the intermediate states, and
you can think about it like that, so long as you remember it is all in your mind.
2. The |`i's are a nice mathematical trick we inserted to solve a huge integral, which is anyway only the
second-order term of the exact expression of the propagator. They are an extremely useful tool, and we
will use Feynman diagrams to represent them, but they do not mean anything real, in the sense that
there is no way to determine whether the system actually goes through |`i.
I like this interpretation.
I like you. Try taking Quantum information Theory in the next semester. I promise you will learn more
about the meaning of reality there than in any philosophy class.
Now we are going to apply this elegant theory to the case of light-matter interaction (Section 2.6.2 in the OS).
By the way, the OS' approximations are not so obvious, so we will do it in a slightly dierent way.
In Section 2.6.1 we applied PT to rst order in time to the rst and second interaction terms of our
Hamiltonian. Now, we will apply PT to second order to the rst term (Eq. 2.15),
r
e X 2π~
V̂ = Ĥint,1 = −√ ĵ(−k) · e(k, λ) ⊗ âk,λ + ĵ(k) · e∗ (k, λ) ⊗ â†k,λ .
L3 k,λ ωk
(1)
We already know that PT to rst order on this term does not give us any of these transitions, Γi→f = 0, so
all the contributions to the transition rate will come from the second-order term (and higher, but we dismiss
those). From Eq. 2.46, we have
2
(2) 2 X hf |V̂ |`ih`|V̂ |ii
Γi→f ≈ Γi→f = δE(i) − E(f ). (2.49)
~ E(i) − E(`) + i~η
`
The intermediate states |`i have the form |`ip ⊗ |Ñk,λ , . . .i [sorry, I am running out of symbols here. . . ]. Let us
focus on the term
hNk,λ − 1, 1|âq,µ |Ñk,λ , . . .ihÑk,λ , . . . |âq0 ,µ0 |Nk,λ , 0i = 0, because we cannot destroy two photons to go from
the initial to the nal state.
hNk,λ − 1, 1|âq,µ |Ñk,λ , . . .ihÑk,λ , . . . |â†q0 ,µ0 |Nk,λ , 0i = Nk,λ if |Ñk,λ , . . .i = |Nk,λ , 1i, (q0 , µ0 ) = (k0 , λ0 ),
p
0 0
and (q, µ) = (k, λ) ; and 0 otherwise. Here, rst we create a photon in mode (k , λ ) and then destroy a
photon from mode (k, λ).
hNk,λ − 1, 1|â†q,µ |Ñk,λ , . . .ihÑk,λ , . . . |âq0 ,µ0 |Nk,λ , 0i = Nk,λ if |Ñk,λ , . . .i = |Nk,λ − 1, 0i, (q0 , µ0 ) = (k, λ),
p
0 0
and (q, µ) = (k , λ ) ; and 0 otherwise. Here, rst we destroy a photon from mode (k, λ) and then create
0 0
a photon in mode (k , λ ).
hNk,λ − 1, 1|â†q,µ |Ñk,λ , . . .ihÑk,λ , . . . |â†q0 ,µ0 |Nk,λ , 0i = 0, because we cannot create two photons to go from
the initial to the nal state.
Note that we impose no restrictions on the intermediate state of the particle, |`ip Our equation has become
much more managable,
√ r
e 2π~ Nk,λ
hf |V̂ |`ih`|V̂ |ii = − √
c L3 ωk ωk0
h`|ĵ(k0 ) · e∗ (k0 , λ0 )|0i δ(|Ñk,λ , . . .i, |Nk,λ , 1i)
hn|ĵ(−k) · e(k, λ)|`i
,
+ hn|ĵ(k0 ) · e∗ (k0 , λ0 )|`i h`|ĵ(−k) · e(k, λ)|0i δ(|Ñk,λ , . . .i, |Nk,λ − 1, 0i)
(2.50)
2
(2) 2 X hf |V̂ |`ih`|V̂ |ii
Γi→f ≈ Γi→f = δ(E(i) − E(f ))
~ E(i) − E(`) + i~η
`
2
4πe Nk,λ
= δ(~ωk0 + En − E0 − ~ωk )
c2 L3 ωk ωk0
2
X hn|ĵ(−k) · e(k, λ)|`ih`|ĵ(k0 ) · e∗ (k0 , λ0 )|0i hn|ĵ(k0 ) · e∗ (k0 , λ0 )|`ih`|ĵ(−k) · e(k, λ)|0i
+ .
E0 − E` − ~ωk0 + i~η E0 − E` + ~ωk + i~η
` | {z } | {z }
|Ñk,λ ,...i=|Nk,λ ,1i |Ñk,λ ,...i=|Nk,λ −1,0i
(2.51)
Now you can calculate the cross-section. [sorry, running out of time] The total rate of photons detected at a
given angle, by the way, is the sum of the the cross-sections given by all interaction terms, with PT going as far
as we nd reasonable.
The Feynman diagram for this transition is represented in Fig. 2.5b. You get two diagrams, representing the
two terms inside that sum: one diagram for the rst term, where the intermediate state has one less photon in
mode (k, λ), and one diagram for the second term, where the intermediate state has one more photon in mode
(k0 , λ0 ).
Very Important Note: Feynman diagrams represent the matrix elements that appear when we apply PT to
an interaction. They are a simple way to visualize all the ugly maths behind the nal result for transition rates
and cross-sections. They do not represent what happens in reality. As we see in Fig. 2.5a, all that we can
observe in the lab is that the particle absorbed a photon and emitted a dierent photon, going to an excited
state, just like in the previous case (Section 2.6.1). Intermediate states are only in our mind.
2.6. LIGHT SCATTERING RELOADED 25
k0 , λ0
Figure 2.5: Transition obtained from PT to second order on time, applied to the rst interaction
term.
Chapter 6
[This chapter is a teaser for quantum chemistry, and a motivation for computational physics.]
The goal: To understand the behaviour of atoms and ions with several electrons.
The problem: Interaction between the electrons makes things horribly complicated. There is no known analytical solution
for two or more electrons. Current numerical simulations are quite good, matching experimental results,
but we won't go there. Instead, we will look at a few approximations and attempts to solve them.
Notation: Z is number of protons in the nucleus, N the number of electrons. They are the same for atoms, dierent
for ions.
N N
X p̂i X Ze2 X e2
H= − + , (6.1)
i=1
2m i=1 |x̂i | i>j
|x̂i − x̂j |
where e is the electron's charge, and all coordinates are relative to the nucleus. The rst term is just the
kinetic energy of every electron, and the second term is the Coulomb attraction to the protons in the nucleus.
This Hamiltonian is already an approximation: we implicitly assume a point-like nucleus with charge Ze,
instead of considering Z protons with their own wave functions (to say nothing of neutrons). In the last term,
we have the interaction between electrons. We sum over j, i > j to avoid double-counting of pairs. This term
was not present in the hydrogen atom, and makes it impossible to nd an exact analytical solution for the
Hamiltonian (think of the three-body problem in classical mechanics). We will go over some approximations
now, but you have been warned: it ain't pretty.
This is a very old approximation that we keep learning mostly out of historical interest. It is crude and classical,
and doesn't really allow us to compute anything very interesting or correct. However, some of the techniques
used pop up again and again in solid-state physics, for example, so it is worth a look.
We assume,
1. N 1 very large;
2. electrons independent;
4. the density of electrons has spherical symmetry, ρ(r) = ρ(r) (this does not happen in reality: some
spherical harmonics do not have this symmetry);
5. there are no electrons near the nucleus, so ρ(r → 0) = 0 (this comes naturally if you do a full QM analysis
of the problem, but has to be forced by hand in a classical approximation);
6. the electrons only feel ρ(r) and not the other individual electrons (mean-eld approximation; see solid-state
physics)
26
6.1. THOMAS-FERMI APPROXIMATION 27
Now we assume that each electron only feels the scalar potential created by this charge density, and not the
individual electrons. This is the so-called mean-eld approximation. The next step is to use Maxwell's equations
to nd the total potential,
(
4π(−e)ρ(r), |r| > 0
−∇2 φ(r) = (6.3)
Ze, |r| → 0,
which tells us that the potential also has spherical symmetry. We can write the classical Hamiltonian of each
individual electron as
pi
Hi = − eφ(ri ). (6.4)
2m
pi P P
We are looking for the ground state of the global Hamiltonian i 2m − eϕ(ri ). In quantum
Ĥ = i Ĥi =
mechanics we build the ground state of the hydrogen atom by putting two electrons of opposite spin per state,
starting at the state with the lowest energy continuing until we ran out of electrons. The energy of the last
occupied state is called the Fermi level, EF .
In this classical approximation, we do something analogous: we populate the states of phase-space (r, p, s)
with a uniform density of electrons, up to EF . We assume that the energy is degenerate in spin (which
is not true), so we can forget about the degree of freedom s if we multiply the density of (r, p) by two. The
volume of a minimal cell of phase space is (2π~)3 , so we get an electron density in phase space of
(
2 p
(2π~)3 , EF > E(r, p) := 2m − eφ(r)
ρ̃(r, p) = (6.5)
0 EF < E(r, p).
To obtain the density of electrons in space, we integrate over all p,
Z
ρ(r) = d3 pρ̃(r, p) (6.6)
Z
2
= d3 p
(2π~)3 E<EF
Z 2π Z π Z √2m(EF +eφ(r))
2
= dφ dθ sin θ dp p2
(2π~)3 0 0 0
8π 1 3/2
= (2m[EF + eφ(r)]) . (6.7)
(2π~)3 3
Here we used the fact that φ(r) ≥ 0, ∀r > 0 (the potential is negative only in the centre where the protons are,
but there ρ = 0).
Now we found ourselves a system of dierential equations,
(
0, r→0
ρ(r) = (2m[EF +eφ(r)])3/2 (6.8)
3π 2 ~3 , r>0
(
2 −Ze, r → 0,
∇ φ(r) = (6.9)
4πeρ(r), r > 0,
1
which people in the twenties had moderate fun solving. We will not do that because with all those classical
assumptions and approximations, the nal density of electrons has little to do with the real wave-function
density in an atom.
Nevertheless, this approximation was used to try to estimate the size of an atom for the rst time. Size"
was dened as the radius of a ball that contains most of the electron density (sort of like the half-life time for
radioactive elements). They obtained two results: when the percentage of charge inside this ball was set to be
a constant, say 98%, then the radius grew with Z −1/3 . When the percentage changed with Z, like (Z − 1)/Z ,
meaning all the electrons except one are inside the ball", then the radius was constant [there is a mistake in
the original script here!]. In other words: depending on how you measure it, the size of the atom decreases or
stays the same as you increase the number of electrons. This is not complete nonsense, because we are also
increasing the number of protons, and therefore the attraction to the nucleus. But it should be a warning that
we seriously misestimated the interaction between electrons.
1 If you want to solve this, add the normalization condition, 6.2, and nd an expression for EF .
28 CHAPTER 6. ATOMS (AND THE PERIODIC TABLE)
R
ρ
Figure 6.1: [Example: the atom is has 98% of the total electron density inside a radius of R.]
1. The electrons are independent, distinguishable particles; in other words, the global wave function of all
the electrons is a product state
where |ψj ii is the ket corresponding to the wave function ψj , living in Hilbert space Hi .
2. The Pauli principle is introduced by hand by saying that we cannot have more than one electron with the
same wave function ψj .
The next step is to solve the SE under these assumptions, to nd the eective potential and the density of
particles. However, as you may imagine, the results are not very accurate, since we stipulated that electrons
are not fermions.
This is a more reasonable approximation, and we will study it in detail. It is widely used both in solid-state
physics and in quantum chemistry.
Assumptions:
1. The electrons are non-interacting fermions; in other words, the global wave function is the anti-permutation
of a product state,
1 X
|ψi1,2,...,N = √ (−1)|P | |ψP (1) i1 |ψP (2) i2 . . . |ψP (N ) iN (6.11)
N ! P ∈SN
= a†N . . . a†2 a†1 |0i, (6.12)
where |ψj ii is the ket corresponding to the wave function of particle i (includes the spatial wave function
φj (x) and spin sj ). Important remark: In general, the global wave function is the anti-permutation of
an arbitrary state. With this restriction, we are excluding many possible correlations between electrons.
Our primary goal is to nd the atom's ground state. However, in general the ground state may have
correlated electrons, so we will settle for the state of non-interacting electrons (of the form of Eq. 6.12) that
minimizes the atom's energy. Formally, we want to nd the single-particle wave-functions {|ψi i} that minimize
hψ|H|ψi (where |ψi is given by Eq. 6.12). This optimization is subject to a normalization constraint on the
single-particle wave-functions, hψ|ψi − 1 = 0, ∀i. We will solve this using the method of Lagrange multipliers:
we dene the Lagrange function
X
Λ(|ψ1 i, . . . , |ψN i, λ ,...,λ ) = hψ|H|ψi − λi (hψi |ψi i − 1), (6.13)
| 1 {z N}
Lagrange multipliers
i
and try to minimize it, ie, nd the arguments of ∇Λ = 0. Before we do it, though, we have to write the term
hψ|H|ψi as a function of the single-particle wave functions. This will require us to write both H and |ψi in the
second quantization formalism. The state is easy,
where |0i is the vacuum. Now the Hamiltonian. Recall that, in general, we can create a many-fermion operator
from a single-particle operator M,
n o
for some basis {|ψi i}i of the single-particle Hilbert space. Note that the operators aj , a†i act on the global
fermionic state, whileM and {|ψi i} refer to a single particle. This new operator has the following physical
M takes state |ψj i to state |ψi i (the weight of this transtion is given by hψi |M |ψj i). If
interpretation: we see if
that happens, we destroy a particle in state |ψj i and create a particle in state |ψi i, using the fermionic operators.
For two-particle operators, this generalizes to
p̂2 Ze2 1 X † † e2
a†i hψi |
X X †
H= |ψj i aj − ai hψi | |ψj i aj + ai aj hψi |hψj | |ψk i|ψm i ak am . (6.17)
ij
2m ij
|x̂j | 2 |x̂k − x̂m |
ijkm
Now we can compute the energy of this state, hψ|H|ψi, one term at the time.
For the two rst terms, with one-particle operators, we have
X † M
hψi |M |ψj i hψ|a†i aj |ψi =
X X
hψ| ai hψi |M |ψj i aj |ψi = hψi |M |ψi i (6.18)
ij ij
| {z } i=1
=δij Θ(N −i)
N N N N
X −~2 ∇2 X Ze2 X e2 X e2
hψ|H|ψi = hψi | |ψi i − hψi | |ψi i − hψi |hψj | |ψi i|ψj i + hψi |hψj | |ψj i|ψi i.
i=1
2m i=1
|x̂i | i,j=1
|x̂i − x̂j | i,j=1
|x̂j − x̂i |
(6.19)
N 2
X ~ 1
Λ(|ψ1 i, . . . , |ψN i, λ1 , . . . , λN ) = − hψi |∇2 |ψi i + Ze2 hψi | |ψi i
i=1
2m |x̂ i|
N N
X 1 X e2
+ e2 hψj |hψi | |ψj i|ψi i − e2 hψi |hψj | |ψj i|ψi i
j=1
|x̂j − x̂i | j=1
|x̂i − x̂j |
+ λi (hψi |ψi i − 1) , (6.20)
where we rearranged the indices of the two-particle terms for later convenience. We are looking for the arguments
(|ψ1 i, . . . , |ψN i, λ1 , . . . , λN ) that minimize it. We can split the minimization condition in two parts:
30 CHAPTER 6. ATOMS (AND THE PERIODIC TABLE)
^ ∂
Λ = 0 ⇔ ∀i hψi |ψi i = 1, (6.21)
i
∂λi
2. All the {|ψi i} must also be local minima. Let us see what that implies in a more general case. Say we
want to nd a local minimum |φ0 i of a matrix element of the form hφ|A|φi, where A is any Hermitian
operator. This means that if you change the function φ0 even a tiny bit, the value of the matrix element
will increase. We can express that in the following way: if instead of |φ0 i we have |φ0 i + t|ηi, where η is
any other function and t is very small, then the matrix element increases. In other words, the derivative
in order to t of the above expression must be zero for t tiny,
∂
hφ0 | + thη| A |φ0 i + t|ηi = 0, ∀|ηi. (6.22)
∂t t=0
This will give us a very nice expression in the end. Stick by as we compute it,
∂
hφ0 |A|φ0 i + thη|A|φ0 i + thφ0 |A|ηi + t2 hη|A|ηi = 0, ∀|ηi
∂t
t=0
[This is slightly dierent from the way I did it in the lecture. This version is more rigorous, and works for
nite-dimensional Hilbert spaces. If we have innitely many dimensions, the calculation becomes more
delicate (though the result is similar). We won't treat that case here.]
Now we have to reduce our problem to the minimization of terms of the form hψ|A|ψi. That is fairly direct,
as
~2 2 2 1
2m ∇ + Ze |xˆi |
N N 1
+e2 j=1 (hψj | ⊗ I) |x̂j −x̂
P
(|ψj i ⊗ I)
X
Λ=− hψi | i| |ψi i − λi . (6.23)
2 N e2
P
i=1
−e j=1 (I ⊗ hψj |) |x̂i −x̂j | (|ψj i ⊗ I)
+λi
| {z }
A
All the {|ψi i} and {λi } count as independent variables, so for each |ψi i we have to minimize hψi |A|ψi i (the nal
λi is a constant). This gives us the minimization condition
0 = A|ψi i
N N
~2 2 1 X 1 X e2
⇔ λi |ψi i = − ∇ |ψi i − Ze2 |ψi i − e2 hψj | |ψj i|ψi i + e2 hψj | |ψi i|ψj i.
2m |x̂i | j=1
|x̂i − x̂j | j=1
|x̂j − x̂i |
We can expand the inner products of the last two terms. Remember that the single-particle wave-functions
included the spatial wave function and spin, |ψi i = |φi i|si i, where φi (x) depends on the position. We obtain
2
~ 2 2 1
λi |φi i|si i = − ∇ + Ze |φi i|si i
2m i |x̂i |
N Z
2
X
3 ∗ 1
−e d xj φj (xj ) φj (xj ) hsj |sj i |φi i|si i
j=1
|x̂i − xj | | {z }
=1
N Z
X 1
+ e2 d3 xj φ∗j (xj ) φi (xj ) hsj |si i |φj i|si i.
j=1
|x̂i − xj | | {z }
=δsi sj
6.3. HARTREE-FOCK APPROXIMATION 31
The spin component is not doing much there (since none of those operators acts on it). Without it, the equation
looks like
~2 2
1
λi |φi i = − ∇i + Ze2 |φi i
2m |x̂i |
N Z
X 1
− e2 d3 yφ∗j (y) φj (y) |φi i
j=1
|x̂ − y|
N Z
X 1
+ e2 d3 yφ∗j (y) φi (y) |φj i δsi sj .
j=1
|x̂ − y|
If we are interested in the spatial wave function at a particular point x, we get, using y := xj inside the
integrals for simplicity,
~2 2
1
λi φi (x) = − ∇x + Ze2 φi (x)
2m |x|
N Z N Z
X 1 X 1
− e2 d3 yφ∗j (y) φj (y)φi (x) + e2 d3 yφ∗j (y) φi (y)φi (x) δsi sj .
j=1
|x − y| j=1
|x − y|
2 N Z
~ 1 X 1 ∗ ∗
=− ∇2 + Ze2 φi (x) +e2 d3 y −φj (y)φj (y)φi (x) + δsi sj φj (y)φi (y)φi (x) .
2m x |x| j=1
|x − y| | {z } | {z }
Hartree term Fock term
| {z }
single particle
This equation looks a lot like a SE, where λi acts almost as an energy eigenvalue. Now we have a system of DE
which can be solved numerically, for instance using as initial {λi , |ψi i}i the energy eigenvalues and eigenstates
of a single-electron atom. Once we have the non-correlated state that minimizes the Hamiltonian, we can apply
small numerical perturbations to that state and see if the energy decreases. The perturbated state will in general
have correlations between electrons.
[See in the script/wikipedia the accuracy of this approximation.]
Chapter 7
I am going to review some basic concepts of quantum mechanics. Bear with me, it will become interesting very
soon. For more details, see the Quantum Information Theory script, by Profs. Renato Renner and Matthias
Christandl,
http://www.itp.phys.ethz.ch/education/lectures_hs11/qit/resources/script
(beginning of chapter 4).
Say you have a quantum state |ψi on a Hilbert space H. We saw in QM1 that quantum states evolve under
unitary operators, |ψ(t)i = Ut |ψ(0)i, with Ut = e−iHt .
There are a few conventions to describe measurements performed on quantum states. A special case is a
measurement where your device can distinguish all the elements of a basis of H. For instance, you measure a
qubit in the computational basis, and your machine gives you two outcomes: 0 when it measures |0i, and 1
when it measures |1i. The probability of obtaining outcome x is given by
2
Pr[x]ψ = |hψ|xi| = hψ|xihx|ψi, (7.1)
To see this, recall that the trace is the sum of the diagonal elements of a matrix, with respect to any basis
P {|ii}i ,
Tr(A) = i hi|A|ii. We have
X
Tr(|xihx| |ψihψ|) = hi|xihx|ψihψ|ii
i
X
= hx|ψi hψ|iihi|xi
i
!
X
= hx|ψi|ψi |iihi| hx|
i
1
| {z }
= hx|ψihψ|xi.
Recap: trace. The trace is linear, αTr(A) + β Tr(B) = Tr(αA + βB), cyclic, Tr(ABC) = Tr(CAB), and
invariant under unitary transformations, i.e., basis-independent, Tr(A) = Tr(U AU † ).
After a measurement with outcome x, the state of the system collapses to |xi. 1 We call the operator |xihx|
the projector onto state |xi. We will generalise it in the next section. Before that, a word about vectors and
matrices in nite-dimensional systems.
32
7.2. DENSITY OPERATOR 33
N
Pick a basis {φi }i=1 of your Hilbert space. We identify kets with the elements of the Hilbert space, for
instance |ii : c 7→ xφi . Now any ket |ψi can be expanded in this basis:
X
|ψi = hφi , ψi |ii, (7.3)
i
hφ0 , ψi
hφ1 , ψi
|ψi = , (7.4)
.
.
.
hφN , ψi
such that row i corresponds to ket |ii. What about bras? They are simply horizontal vectors,
This is compatible with our previous denition of bra as a function that takes a wave function and returns an
inner product. In fact,
hφ0 , ψi
hφ1 , ψi
hτ | |ψi = (hτ, ψ0 i, hτ, φ1 i, · · · , hτ, φN i)
.
.
.
hφN , ψi
X
= hτ, φi i hφi , ψi
i
= hτ, ψi,
hence we can write inner products as contractions between bras and kets, hτ, ψi = hτ | |ψi =: hτ |ψi. Similarly,
operators that act on the Hilbert space (endomorphisms on H) can be represented as matrices,
!
X X
A= |iihi| A |jihj|
i j
1
| {z }
1
| {z }
X
= hi|A|ji |iihj|
| {z }
i,j
hφi ,Aφj i
a11 a12 ··· a1N
a21 a22
= . , aij = hi|A|ji. (7.6)
..
.. .
aN 1 aN N
Examples. Say that H is the Hilbert space of a qubit. Here go a few quick examples of kets, bras, and
operators in the computational basis, {|0i, |1i}.
1 0 0 0 0 1
|0i = , |1i = , |1ih1| = , |0ih1| = ,
0 1 0 1 0 0
|0i − |1i 1 1 |0i − |1i h0| − h1| 1 1 −1
|−i = √ =√ , |−ih−| = √ √ =
2 2 −1 2 2 2 −1 1
h i 0 0 1 1 −1 1 0 0 1
Pr[1]|−i = Tr |1ih1| |−ih−| = Tr = Tr = .
0 1 2 −1 1 2 −1 1 2
Imagine that you are given a quantum state to play with. The only problem is that you are not sure what state.
For instance, you may have a awed source of quantum states, which was supposed to produce state |ψi, but
34 CHAPTER 7. STATISTICAL QUANTUM MECHANICS
will produce state |τ i instead with probability p. What is the probability of obtaining |xi when you measure
your unknown state in that basis? Well, it has to be [probability of having |ψi] ∗ [probability of measuring |xi
on |ψi ] + [probability of having |τ i] ∗ [probability of measuring |xi on |τ i ],
Here, we dened the density operator ρ = (1 − p) |ψihψ| + p |τ ihτ |. This is a mathematical object that reects
our ignorance about the actual state of the system. For innite dimensions, it becomes a matrix (the density
matrix). For instance, if |ψi = |1i and |τ i = |−i, we have
ρ = (1 − p) |1ih1| + p |−ih−|
p
− p2
0 0 1 1 −1
= (1 − p) +p = 2 .
0 1 2 −1 1 − p2 1 − p2
The probability of obtaining a 1 when measuring this state in the computational basis is
Pr(1)ρ = Tr[|1ih1| ρ]
p
− p2
0 0 2
= Tr
0 1 − 2 1 − p2
p
0 0 p
= Tr =1− .
− p2 1 − p2 2
Note that we could also have computed this quantity the old way,
We have just showed that the time evolution of a density operator is simply given by
Now, this is more relevant than it might appear at rst sight. Think about what you can do in a lab: you can
let systems evolve, by manipulating Hamiltonians, and you can measure them. You have showed that both time
evolution and measurement statistics of a system depend only on the density operator (and the Hamiltonian).
This implies that it is impossible to distinguish two systems with the same density operator, even if they were
produced in dierent ways. Let me give you a somewhat trivial example: suppose that your friend Alice has
two machines that produce polarized photons. The only problem is that those machines are not reliable at all:
machine A produces state |0i with probability 1/2 and state |1i with probability 1/2, while machine B produces
B produces state |+i with probability 1/2 and state |−i with probability 1/2. The density operators of a state
coming from each machine are
1
1 1 1 1 0 1 0 0
ρA = |0ih0| + |1ih1| = + = ,
2 2 2 0 0 2 0 1 2
1
1 1 1 1
1 1 1 1 −
ρB = |+ih+| + |+ih+| = 2
1
2
1 + 2
1 1
2 = .
2 2 2 2 2 2 − 2 2 2
7.2. DENSITY OPERATOR 35
Now imagine that Alice gives you a million photons, and tells you that she used the same machine to produce all
of them. But she won't tell you which machine. You would think that it would be easy to nd out, right? Here
are a million photons, either half are |0i and half are |1i or half are |+i and half are |−i. You can do whatever
you want to your photons: rotate the states, measure them, let them evolve under convoluted Hamiltonians,
make them interact with an external system, let your local soothsayer examine them. And yet you will never be
able to determine which machine was used, because photons from A and from B have the same density matrix.
2
The origins of a quantum state do not matter.
X
ρ= pi |ψi ihψi |, (7.10)
i
P
where the {pi } form a probability distribution (meaning ∀pi ≥ 0 and i pi = 1), and the {|ψi i} are
quantum states. This has the following physical interpretation: the system is in state |ψi i with probability
pi .
P
By denition, the {pi } are the eigenvalues of ρ, therefore Tr(ρ) = i pi = 1, ρ is positive semi-denite,
and Hermitian.
A system is said to be in a pure state if the corresponding density operator only has one non-zero eigenvalue,
{pi } = {1, 0, , 0 . . .}, and therefore ρ = |ψihψ|. Otherwise a state is considered mixed (meaning that there
is more than one possibility for the exact state of the system).
If ρ = 1/N , where N is the dimension of the Hilbert space of interest, it is called fully mixed.
In a nutshell, the dierence is that the former is a pure state (you know that the system is in that exact state),
while the latter is a probabilistic mixture of two possible states. To clarify things, we can write the density
matrices corresponding to both states, in two dierent bases:
1 1
1 0 1
Pr(0)σ = Tr[ |0ih0| σ] = Tr 2
1
2
1 , =
0 0 2 2 2
1
1 0 0 1
Pr(0)ρ = Tr[ |0ih0| ρ] = Tr 2
1 = .
0 0 0 2 2
This does not look very promising: we could not distinguish the two states solely from the statistics of this
measurement. But what happens if instead we measure them in basis {|+i, |−i}? The probability of obtaining
|1ih1|ρ|1ih1| + |2ih2|ρ|2ih2|
=
Tr(P< ρ)
P< ρP<
= .
Tr(P< ρ)
This generalizes to any projective measurement with outcomes {i} and corresponding projectors {Pi },
Pi ρPi
Pr(i)ρ = Tr(Pr ρ), ρi = .
i Tr(Pri ρ)
For further generalizations. . . I can't emphasize it enough, come to the Quantum Information Theory next
semester!