Lecture 11: Time-Dependent Perturbation Theory: Harmonic Perturbations (10/25/2005)

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Lecture 11: Time-dependent perturbation

theory: harmonic perturbations (10/25/2005)


Last time we started time-dependent perturbation theory. We understood
how the approximate results may be calculated by iterations; the method
is related to another method, the Born approximation, one that we will
discuss later in the course. At the very end, we considered the effect of
time-dependent perturbations that factorize into h0 (r)f (t) and calculated
the result for f (t) being a pulse (constant in an internal, otherwise zero).
Additional reading if you wish: Griffiths ch. 9.2
However, there is one more function f (t) of time that is perhaps even more
important: the harmonic function f (t) = cos(t). The adjective harmonic
always represents a (co)sinusoidal dependence. For example, a harmonic
oscillator has x(t) that is proportional to cos(t), too.
More generally, the word harmonic describes a function whose Laplacian (or
second derivative) is a multiple of itself but in this class, we wont use the word
harmonic in this more general sense.

Harmonic time-dependent perturbations


So what are the results for these perturbations? They are of the form
0 (r, t) = V (r) cos(t)
H

0
Hba
(t) = Vba (r) cos(t)

Recall how the algorithm from the previous lecture worked: you start with
a state at t = 0, let us call it |ai, and the formula for the first iteration
approximating the amplitude to be in another state |bi at time t is
(1)

t
1
1
0
Vba
cos(t0)eiba t dt0 = . . .
ba (Eb Ea )
i
h
h

0
Z t

1
=
ei(ba +)t + ei(ba )t dt0
Vba
2i
h " 0
#
Vba ei(ba +)t 1 ei(ba )t 1
=
+
h

ba +
ba

cb (t) =

If Eb > Ea i.e. ba > 0, the second term gets large for ba . Similarly,
for Ea > Eb i.e. ba < 0, the first term gets large for ba .
1

Without a loss of generality, let us assume Eb > Ea . Let us study


the neighborhood of the right frequency where we can drop the first term.
This approximation is called the rotating wave approximation because it is
equivalent to starting with one of the two complex terms only, namely with
1
H 0 (r, t) = V (r)eit .
2
Fine. So we only have the second term, but its numerator is still a difference
of two terms. Again, we factorize out the average phase:
(1)
cb


Vba ei(ba )t/2  i(ba )t/2
e
ei(ba )t/2 =
=
2
h ba
Vba i(ba )t/2 sin[(ba )t/2]
= e
h

ba

This is the amplitude and the probability is obtained by squaring its absolute
value:
|Vba |2 sin2 [(ba )t/2]
(1)
.
Pab (t) |cb |2 =
(ba )2
h
2
Now we must draw a graph of this function as a function of . The maximum
is achieved for ba = and it is
Pmax =

|Vba |2 t2
h
2 4

which implies that the probability of the transition increases with time quadratically if the frequency is exactly adjusted. Note that eventually Pmax > 1
which does not mean that probabilities can exceed 100 percent and mathematics collapses; instead, it means that the first-order approximation breaks
down.
Otherwise, the probability oscillates between zero and an upper bound of
const/(ba )2 where const is independent of time. It is really interesting
that the probability of transition is zero for
t=

2k
,
|ba |

k = 1, 2, 3, . . .

The width of the peak near ba is inversely proportional to t which is


related to the Et h
uncertainty relation. The more time we dedicate
to measure the resonant frequency, the more accurately we can determine it.
How can you interpret the transition from |ai to |bi at the resonant frequency = ba ? You may view the perturbation to be a beam of quanta
whose energies are h
, and if the frequency is the right one, your system can
absorb this energy h
ba and jump from |ai to |bi. But you see that it works
both ways; the same external perturbation also encourages our system to
drop from |bi to |ai an example of stimulated emission that we will discuss
later.
2

Avoiding iterations: exact results


So far we have calculated the transition probabilities using the first-order
iteration. Actually it is possible to calculate the exact result, at least in
the rotating wave approximation (which is independent from the interative
approximation). It may be fun to solve the problem exactly to see how exact
the first iteration was. The relevant equations are
1
ca
=
Vab ei(ba )t cb (t)
t
2i
h
cb
1
=
Vba e+i(ba )t ca (t),
t
2i
h

Vba = Vab

Now we want to eliminate ca (t). To do so, it is helpful to first differentiate


the second equation.
1
ca
2 cb
=
Vba e+i(ba )t i(ba )ca (t) +
2
t
2i
h
t
"

That was helpful because now we may substitute both for the first term in
[. . .] which is proportional to ca (t) from the original second equation as
well as for its time-derivative from the original first equation. We obtain
2 cb
cb |Vba |2
cb (t)
=
i(

ba
t2
t
4
h2
and ca (t) is gone and we recognize an equation in which the middle term is a
friction term. Moreover, linear differential equations with fixed coefficients
have solutions of the type
cb (t) = A exp(t).
Substituting this Ansatz into the second-order differential equation, we obtain after we divide by cb (t) itself:
2 i(ba ) +
=

|Vba |2
=0
4
h2

q
1
i(ba ) (ba )2 |Vba |2 /
h2
2


Let us now celebrate Isidor Rabi and use the name Rabi flopping frequency
for an ingredient in the solution we just found:
R

1q
+(ba )2 + |Vba |2 /
h2 .
2

We will explain the word flopping momentarily. At any rate, the new
notation allows us to write the general solution as


cb (t) = ei(ba )t/2 A e+iR t + B eiR t

Imposing boundary conditions


At time t = 0, we want the state to be purely |ai. This means
cb (0) = 0.
In other words, looking at the general solution,
A = B.
The condition ca (0) = 1 means
Vba
ba
cb
|t=0 =
=
[A A] + iR [2A]
t
2i
h
2i

A=

Vba
4
h R

The full cb amplitude is thus


cb (t) =

Vba i(ba )t/2


sin(R t)
e
2
h R

and the transition amplitude simply


Pab

Vba 2 2
sin (R t).
= |cb (t)| =
2
hR
2

This formula actually looks simpler than it is because we have used the
symbol R . If we expand it, the perturbative vs. exact answers look a bit
different; spot the differences:
Pab =

|Vba |2 sin2 [(ba )t/2]


,
(ba )2
h
2
q

2
h2 t/2]
|Vba | sin [ (ba )2 + |Vba |2 /
=
,
h
2
(ba )2 + |Vba |2 /
h2
2

Pab

perturbative
exact

You see that the two answers agree if


either t  1/(ba ): you may then use sin X X in both formulae
or |Vba |2 /
h2  (ba )2 : you neglect the Vba -like terms in the ratio
At any rate, neither of these two conditions may be satisfied for = ba ,
on resonance not even if Vba is tiny. In this case
R =

Vba
2
h

Pab = sin2 (R t)

You see that no matter how small Vba is, the probability will reach 100 percent
by t = /2R . The physical system oscillates between two states and between
emitting and absorbing a quantum of energy (a photon, for example).
4

The width
Let us continue to look at the exact answer. We have discussed the resonance.
But what about 6= ba ? Let us choose, for example,
|ba | =
Then

where R0

Vba
.
h

2Vba
1q
2
= 2R0
(ba )2 + |Vba |2 /
R =
h =
2
2
h
is R at resonance. The probability
Pab =

1 2
sin ( 2R0 t)
2

oscillates between zero and one-half. Now imagine you try a more general
|ba | and draw the maximal transition probability. You will obtain a
kind of Laffer curve. Draw it! Its maximum in the middle is at Pab = 1
and you obtain Pab = 1/2 for
ba = |Vba |/
h.
Consequently, the width of the bell curve measured in the middle is 2|Vba |/
h.
You see that the resonance width narrows if Vba decreases. The shape of the
curve will reflect the function 1/(1 + x2 ) the general simplified form of the
exact formula for the probability.

Spin magnetic resonance


and nuclear magnetic resonance (NMR)
These two phenomena (more precisely one) provide us with a great opportunity to apply the insights we have just found. First, let us put a generic
spin-1/2 particle in the magnetic field in the z-direction. Its Hamiltonian is
= ~ B
~ = S
~ B
~ = Sz Bz
H
where is the gyromagnetic ratio
g

q
2m

where g is about two for a point-like fermion but can have other values for
nucleons and nuclei. For example, g 5.59 for a proton. The Hamiltonian
is therefore
!

h
+1
0
H = Bz
0 1
2
and the states up i.e. (1, 0)T and down i.e. (0, 1)T are eigenstates of the
energy with E =
hBz . As we discussed in the previous lecture, these
5

energy eigenstates are stationary but a spin in the xy-plane will precess with
a frequency
E
= Bz .
=
h

But we really want to study time-dependent perturbations. An extra magnetic field in the z-direction would not induce any transitions. We want to
choose a different direction for the additional magnetic field, namely the xdirection (without a loss of generality). In classical physics, such an extra
field would apply a torque that would however be cancelled on the opposite
sides of the classical orbit. The effect would be nothing else than a slight tilt
of the orbit.
However, if the Bx field depended on time sinusoidally with the orbital
frequency, the torque would tend to be in the same direction and it would
attempt to flip the spin.
NMR technique
The technique for NMR is therefore to have a large B-field in the z-direction
and a small oscillating B-field in the x-direction. The full Hamiltonian is
therefore
= H0 + H 0 = [Sz Bz + Sx Bx cos(t)]
H
"
!
!#
h

+1
0
0 +1
+ Bx cos(t)
= Bz
0 1
+1
0
2
But it is a special case of the general problem we solved 5 minutes ago. The
resonance appears at ba i.e. at
= Bz ,
the same frequency as the classical frequency of precession; note that even
though the formula depends on Bz , it is the required frequency of the Bx perturbation. The corresponding frequency R see the definition is |Vba |/2
h
at resonance which means
Bx
R =
4
where one factor 1/2 came from the beginning of the Hamiltonian and the
other one from the formula for R . At resonance, the spin flip1 probability
will be


Bx t
P = sin2
4
If there were an equal number of spins in the up direction as spins in the
down direction, the flips would cancel out and no signal would be observed.
However, the |upi and |downi states have a different energy and there must
1

Now you see why we called R the Rabi flopping frequency.

be an imbalance governed by the Boltzmann distribution. Consequently, a


pick-up coil sensitive to Bz will see a signal at radio frequency (rf) R as
some of the spins migrate back and forth between |upi and |downi.
Practical applications
Note that the resonant frequency and R depend on . The appropriate
values of of the nuclei in your sample may therefore be measured; this
is referred to as NMR spectroscopy. NMR may be used for medical imaging. Recently, the abbreviation NMR is replaced by MRI (magnetic resonance imaging) or MRT (magnetic resonance tomography) in this context.
Tomography is able to measure the concentration of various isotopes in twodimensional slices of the patients body.
NMR is also used for accurate measurements of the magnetic fields because the resonance frequency may be measured very accurately.

Ammonia masers
A week ago, we discussed the molecule of NH3 as an example of a two-level
system. The nitrogen may be either above or below the plane of the triangle
made out of the three hydrogen atoms. However, the energy eigenvectors are
the symmetric or antisymmetric combinations of these two states, and their
energies may be called E0 A.
For our present purposes, it is also important that NH3 has a permanent
electric dipole moment perpendicular to the hydrogen plane whose sign is
correlated with the position of the nitrogen atom. In the electric field E,
the molecule gains or loses the energy E. This term in the energy would
be proportional to diag(1, 1), but if we choose the energy eigenstates as
our basis, it will be rather proportional to the Pauli matrix x . The full
Hamiltonian is
= H0 + H 0 = E0
H

1 0
0 1

1 0
0 1

+ E0 cos(t)

0 1
1 0

The term E0 is just an irrelevant additive shift to the energy and may be
dropped. The rest of the Hamiltonian is mathematically isomorphic to the
spin magnetic resonance case. We may immediately see that
=

2A
,
h

R =

E0
.
2
h

At the resonant frequency = 2 24 GHz, the ammonia molecule will


be flipping between the symmetric and the antisymmetric state, with the
probability being
Pab = sin2 (R t).

If all molecules start in the antisymmetric state at t = 0, by the moment


t=

h
=
2R
2E0

all of them flip to the symmetric state which may be interpreted as the
emission of one photon of energy 2A per molecule. Even a small signal
E0 will be amplified in this way, after a long enough time. We have just
(theoretically) constructed a
MASER = Microwave Amplification by Stimulated Emission of Radiation.
It differs from a LASER in the same way as might (m) differs from light (l).
To construct it in practice, we must also be able to bring all the molecules to
the antisymmetric, higher-energy state. This goal, as Feynman explains in
his volume III, may be achieved by a splitting strategy in which the molecules
are sent to a non-uniform electric field. Because of their electric dipole, the
NH3 molecule beam will split into two beams. We will only use one of them.
All molecules in this beam are in a well-defined state, and by acting on it
properly with the electric field in an orthogonal direction, we may bring them
to the antisymmetric state.
(Draw a beam split to two beams I,II by an electric field. Draw a cavity
of the proper length vT where v is the velocity and T is the right time that
transforms all incoming II molecules to I.)

If the antisymmetric molecules are sent into a resonant cavity of a proper


length, each molecule delivers the energy 2A to the cavity. By reducing
the driving (oscillating) field, we may make the width of the resonance very
narrow. MASERs may be used for very precise atomic clocks based on their
natural, characteristic frequency .

You might also like