Chapter 2. 1D Wave Mechanics: X P I X X X X

Quantum Mechanics
Chapter 2. 1D Wave Mechanics
In this chapter, we discuss the simplest and conceptually most important 1D problems of wave
mechanics, including several issues which are not typically covered in undergraduate courses but are
crucial for our discussion of more advanced topics in the following chapters.
2.1. Free particle: Wave packets

As was discussed in the end of Chapter 1, in several cases the general (3D) Schrodinger equation
(1.52) may be reduced to the 1D equation for the x-component of the wavefunction:
 ( x, t )  2  2  ( x, t )
i   U ( x )  ( x, t ) . (2.1)
t 2m x 2
Let me repeat that in most cases the effective 1D potential U(x) differs from the genuine potential
energy by some constant. Although this constant may be readily absorbed into the particle energy (see
the next section), this fact is still important to remember.
Let us start with the discussion of unconstrained 1D motion, with U(x) = 0. From what we have
done in Chapter 1, it is clear that in this case the fundamental solution of the stationary Schrödinger
equation (1.47) is a monochromatic wave
i  p2
 ( x, t )  const  exp  p 0 x  E ( p0 )t  , with E ( p 0 )  0 . (2.2)
  2m
Equation (1.22) tells us that since for this wavefunction, product * does not depend on either x or t,
the particle is completely delocalized, i.e. its probability is spread all over axis x, at all times. (As a
result, such state is still compatible with the uncertainty relation (1.34), despite the exact value p0 of
momentum p. 1)
Now let us form, at the initial moment of time (t = 0), a wave packet of the type shown in Fig.
1.6, with a finite “width” (coordinate uncertainty) x, by multiplying wave (2) by some smooth
“envelope” function of x. For example, consider a Gaussian packet
1  x2  i 
 ( x ,0 )  exp - 2 
 exp p 0 x  . (2.3)
(2 ) (x)
1/4 1/2
 (2x)   
(Actually, exactly this packet is shown in Fig. 1.6a.) The pre-exponential factor is this formula has been
selected in the way to have the probability density
1  x2 
w( x, t )   * ( x,0) ( x,0)  exp , (2.4)
2 1 / 2 x  2(x) 2 
1 In this chapter, we will drop index x in the expression px, for brevity.
© K. Likharev, 2008 1
Quantum Mechanics
normalized to P = 1, for any parameters x and p0.2

In order to explore the evolution of this packet in time, we could try to solve Eq. (1) with the
initial condition (3) directly, but according to our discussion in Sec. 1.4, it is easier to proceed
differently. Let us first present the initial wavefunction (3) as a sum (1.50) of eigenfunctions p(x) of
the corresponding stationary 1D Schrödinger equation (1.54) which, according to Eq. (2), are just
monochromatic waves
i 
 p  c p exp px  , (2.5)
 
As was already discussed in Sec. 1.4, we are using index p rather than n to register the fact that, in
contrast with the case of confined motion, the possible values of momentum p form a continuum rather
than a distinct set.3 As the result, the sum (1.50) is turning into an integral:4
 x
 ( x,0)   c p p ( x)dp   c p expip dp , (2.6)
 
where we have included the constant multiplier of Eq. (5), into function c(p). Equation (6) is of course
just the usual Fourier transform, and we can use the well-known formula of the reciprocal transform
1  x  x  1 1  x2 ( p  p) x 
cp 
2   ( x , 0 ) exp 

 ip  d   
     2 (2 ) (x)
1/4 1/2 
exp -
 (2x)
2
i 0

dx

(2.7)
This integral may be worked out by complementing the exponent to the full square of a linear
combination of x and (p0 - p), plus a term independent of x:
2
x2 ( p  p) x 1  2i (x) 2 ( p  p 0 )  (x) 2 ( p  p 0 ) 2
- i 0  x    . (2.8)
(2x) 2
 (2x) 2    2
 
Since integral (7) should be taken at constant p, in infinite limits, it does not change if we replace dx by
dx’  d[x + 2i(x)2(p – p0)/]. As a result, for cp we also get the Gaussian distribution, centered to p0
(Fig. 1.6b):

2 The corresponding table integral,
 exp{ }d   , is very common, and worthy of remembering.
2


 exp{ 2 }d   / 2 , may be readily used to prove that the r.m.s.

2
Another table integral of the same type,

uncertainty of x in the wave packet (3), defined by Eqs. (1.32) and (1.33), indeed equals x, thus justifying that
notation. (The average x is evidently zero, due to the symmetry of wave packet (3) about point x = 0.)
3 Another difference between that case and the free particle (which may be considered as an infinitely-wide
potential well (1.53)) is that in the latter case the traveling waves (5) form a more natural eigenfunction basis.
This difference should not bother the reader too much, because the standing-wave eigenfunction (55) may be
presented as a sum of two traveling waves (5), with equal and opposite values of p, and hence the same
eigenenergy E = p2/2m.
4 This integral, just as all other integrals of this section, has infinite limits [-, +]. For brevity, we will drop the
limits in all formulas below.
Quantum Mechanics
1 1  (x) 2 2  x' 2 
cp  exp   ( p  p )  exp - 2 
dx'
2 (2 )1/4 (x)1/2
0
 (2x) 
2
  
1/ 2
(2.9)
 1  1  ( p  p0 ) 2 
  exp  2 
.
 2  (2 )1/4 (p)1 / 2  (2p) 
with the constant p defined as

p  . (2.10)
2x
Thus we can present the initial wave packet as
1/ 2
 1  1  ( p  p0 ) 2   x
1/ 2 
 ( x,0)    exp 2 
expip dp . (2.11)
 2  (2 ) (p )  (2p) 
1/4
 
It is straightforward to use Eqs. (1.23), (1.26), (1.32) and (1.33) to check that the r.m.s. uncertainty of
momentum in this packet is indeed equal to p defined by Eq. (10), thus justifying the notation.
Comparison of that equation with Eq. (1.34) shows that the Gaussian packet presents the ultimate case
when the product xp has the lowest possible value (½); for any other envelope shape the uncertainty
product is larger. We could of course get the same result for p from Eq. (3) of the initial packet; the
real advantage of Eq. (11) is that it can be readily generalized to t > 0. Indeed, we already know that the
time evolution of the wavefunction is given by Eq. (1.51); for our case giving
1/ 2
 1  1  ( p  p0 ) 2   x  p2 
(2 )1/4 (p)1 / 2 
 ( x, t )    exp   2 
exp ip  exp   i t dp . (2.12)
 2   ( 2p )      2 m 
Since the last exponent is purely imaginary, the extension p of the packet in the momentum space does
not change with time. This is not true for it spatial expansion. In order to show that we need to work out
integral (12) to get an explicit expression for (x,t). This can be done by using the same method as
above, i.e. merging the exponents under the integral, and presenting them as a full square of linear
combination of x and p:
 x  x0 (t ) ip 0 x ip 02 t
2 2
( p  p0 ) 2 x p2 (t )  i
  ip  i t  ( p  p 0 )  x  x 0 (t )     , (2.13)
(2p) 2  2m 2  2(t )  4(t )  2m
where we have introduced two functions of time,
p0 t it
x0 (t )  , and (t )  (x) 2  , (2.14)
m 2m
and used relation (10) between p and x. Integrating over p, we get
1/ 2
1  x   [ x  x0 (t )] 2 i  p 2 t 
 ( x, t )    exp   p 0 x  0  . (2.15)
(2 )1 / 4  (t )   4(t )  2m 
The imaginary part of the ratio 1/ in the exponent gives just an additional contribution to wave’s phase,
and does not affect the resulting probability distribution
Quantum Mechanics
x  ( x  x0 ) 2 1 
w( x, t )   *  exp Re . (2.16)
2 1 / 2 (t )  2 (t ) 
This is the standard Gaussian bell curve, centered to point <x> = x0(t), with the r.m.s. deviation
1 2
  1  (t )  t  1
2
x(t ) 2
 Re     x 
2
  (2.17)
  (t )   x 2  2m  (x)
2
from the center. Figure 1 shows several snapshots of the real part of the wavefunction, for a particular
case p0 = 10p = 5/x.
1 t0 x x
t4 t  10
v0 v0
Re  0
)
1
Fig. 2.1. 1D wave
packet spreading.
 10  5 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70
x / x
We see that the packet moves along x axis, simultaneously broadening (“spreading”) with time.
The first of these results is natural: in any corpuscular theory, a free particle should move “ballistically”
as a whole with the initial velocity v0 = p0/m. On the other hand, in the general wave theory, a narrow
wave packed should move with the “group velocity” vgr = d/dk, with the derivative calculated in the
center of the wave number distribution. Let us see what it gives in our case:
d d ( E / ) dE d ( p 2 / 2m) p0
vgr  k  k0  p  p0  p  p0  p  p0  . (2.18)
dk d ( p / ) dp dp m
We see that (very fortunately!) the wave packet speed equals to that of the classical particle, in
accordance with the correspondence principle. This was the central point of L. de Broglie’s ideas. Just
two remarks are appropriate here:
(i) Notice that this is the group velocity (rather than the phase velocity vph  [/k]k=k0 = p0/2m)
which corresponds to the classical speed of the particle.
(ii) In the general wave theory, the group velocity definition is only valid for narrow wave
packets (k << k0), and for any nonvanishing width, the first of Eq. (18) has corrections proportional to
higher derivatives dn/dkn of the media dispersion law (k). In the particular case of non-relativistic
Schrödinger equation, the dispersion relation (1.29) is quadratic, so that all higher derivatives (with n >
2) vanish, so that result (18) is exact.
The second effect, the wave packet’s spreading with time, is described by Eq. (17). This effect is
also known from the general wave theory, where it is determined by the second derivative d2/dk2 of the
dispersion law, which in our case is constant and equal to /m – the coefficient clearly visible in the
second term of Eq. (17). Physics of the spreading is very simple: for any finite width p of the
Quantum Mechanics
momentum distribution, the genuine group velocity vgr = p/m of each small fraction dp of the component
monochromatic waves is different, resulting in the gradual accumulation of the differences of the
distances traveled by different groups. The most curious feature of Eq. (17) is that the packet width at t
> 0 depends in a non-monotonic way from its initial width x = x(0), and tends to infinity both at
x(0)→ 0 and x(0)→ ∞.5 The physics of such dependence is clear from the above discussion: if we try
to decrease x(t) by compressing the initial packet, by the virtue of relation (10) we increase width p of
its momentum distribution, and hence the speed with which various groups of monochromatic waves run
away from each other at large times.
Notice that for any t  0, the wave packet retains its Gaussian envelope, but the ultimate relation
(10) is not satisfied, px(t) > /2 (due to a nonlinear phase shift between the component waves). The
same is also true for any wave packet whose shape differs from the Gaussian form.
In order to discuss such general case, let us generalize our result (12) to an arbitrary initial time
t0 < t, and an arbitrary initial wave packet shape (x,t0):
 px   p2 
 ( x, t )   c p expi  exp i (t  t 0 )dp , (2.19)
    2m 
with cp described by the first of Eqs. (7). (Equation (19) is of course just an integral form of Eq. (1.51),
for the particular form (5) of eigenfunctions , and eigenenergies E(p).) Combining these equations, we
can write our result in a single line:
1  px0   px   p2 
 ( x, t ) 
2   dp  dx 0  ( x ,
0 0t ) exp 

 i 
 
exp i
  
 exp   i
 2m
(t  t 0 ).

(2.20)
After changing the order of integration, this expression may be evidently rewritten as
 ( x, t )    ( x0 , t 0 )G ( x, t ; x0 , t 0 )dx0 , (2.21)
where function G(x,t;x0,t0) is called the quantum-mechanical propagator.6 For a free 1D particle, this
function,
1  p( x  x0 )   p 2 (t  t 0 ) 
2  
G ( x, t ; x 0 , t 0 )  expi  exp i dp , (2.22)
    2m 
may be readily calculated explicitly using the same trick of complementing the exponent to the full
square. The result is
1/ 2
 m   m( x  x 0 ) 2 
G ( x, t ; x0 , t 0 )    exp . (2.23)
 2i(t  t 0 )   2i (t  t 0 ) 
Notice the following features of this function:
5 For a fixed t, one can find the compromise value x(0) with minimizes x(t).
6 The notation stems from the fact that the wave-mechanical propagator is essentially the Green’s function of the
Schrödinger equation, very similar to those used in other ordinary and partial differential equations of
mathematical physics – see, e.g., CM Sec. 4.1 and/or EM Sec. 2.6 and 7.3.
Quantum Mechanics
(i) It depends only on differences (x - x0) and (t – t0). This is natural, because the problem is
uniform (translation-invariant) both in space and time, i.e. does not have any special features on axes x
or t.
(ii) The function shape does not depend on its arguments – they just rescale the same function:
its snapshot (Fig. 2) just becomes broader and lower with time. It is curious that the spatial broadening
scales as (t – t0)1/2 – just at the classical diffusion, paving a smooth crossover between the quantum
mechanics and classical statistics – see Chapter 7.
0.5
Re  G ( x, t ; x 0 , t 0 )
  0
Im  m /  (t  t 0 )
Fig. 2.2. Real (solid line)
and imaginary (dashed line)
 0.5
parts of the free particle’s
 10 0 10 propagator.
( x  x 0 ) /  (t  t 0 ) / m
For a particle moving in an arbitrary potential, the propagator may lack these features (besides
the uniformity in time, which is retained for any time-independent potential); however, the linear
relation (21) still holds, due to the Schrödinger equation’s linearity. The physical sense of the propagator
may be understood by considering the following special initial conditions:7
 ( x0 , t 0 )   ( x0  x' ) , (2.24)
where x’ is a certain point within the domain of particle’s motion. In this particular case, Eq. (21)
evidently gives
 ( x , t )  G ( x, t ; x ' , t 0 ) . (2.25)
Hence, the propagator, considered as a function of x and t only, is just the solution to the Schrödinger
equation with -functional initial conditions or, in physical language, the wavefunction of a particle
which was initially localized in point x’. Thus while Eq. (19) may be understood as a mathematical
expression of the linear superposition principle in momentum (i.e., reciprocal space) domain, Eq. (21)
may be understood as an expression of this principle in the direct space domain: the system’s “response”
(x,t) to an arbitrary initial condition (x0,t0) is just a sum of its responses to its thin spatial “slices”.
The propagator G(x,t;x0,t0) presents the weight of each slice in the sum.
With this understanding on hand, let us return for a second to Fig. 2. According to the
uncertainty relation, the ultimately compressed wave packet (24) has an infinite width of momentum
distribution, and the quasi-sinusoidal tails of the free-particle propagator, clearly visible in Fig. 2, are the
results of the free propagation of the fastest (highest-momentum) components of that distribution, in
both directions from the packet center.8
7 Notice that this is not equivalent to the -functional initial probability density w(x0,t0) = (x0,t02.
8 Equation (21) will be our entry point for the discussion of Feynman’s path-integral picture of quantum
mechanics in Sec. 4 below.
Quantum Mechanics
In the following sections, we will mostly focus on the behavior of monochromatic wavefunctions
(which, for unconfined motion, may be interpreted as wave packets of very large spatial width x), only
rarely coming back to the wave packet discussion. Our excuse is of course the linear superposition
principle, i.e. our conceptual ability to restore the general solution from that of monochromatic waves of
all possible energies. However, the reader should not forget that, as the above discussion has illustrated,
mathematically this restoration is not always easy.
2.2. Wave mechanics in piecewise-constant potential profiles

Now, let us proceed to the cases in which the potential energy U(x) (including the energy of
quantum confinement in two other directions – see Sec. 1.6) is not identically equal to zero. The easiest
case, of course, is when U(x) = U0 = const. Indeed, in this case the eigenstates of the Schrödinger
equation (1.48), are the same monochromatic waves as for a free particle, just with
p2
Ep  U0 . (2.26)
2m
This means that the only change of the wavefunction is an additional phase factor exp{-iU0t/). Looking
at the basic equations (1.22) and (1.23), it seems that this additional factor does not affect the particle
probability distribution, or any observable, and hence the frequency associated with U0 is a
mathematical artifact. This is certainly true for a single particle,9 however the situation changes as soon
as we recall that the Universe consists of more than one particle.
For example, consider two similar particles, each in the same (say, ground) eigenstate, but with
the potential energies (and hence eigenenergies E1,2) different by a constant U0. Then, presenting their
wavefunctions in the polar form (1.67), we see that the difference   1 - 2 evolves in time as
U0
 t  const . (2.27)

Again, if the particles do not interact, this evolution is unobservable; however, a very weak coupling
may allow to observe the process, while keeping the particle dynamics virtually unperturbed, so that Eq.
(27) is still valid.
Probably the most dramatic demonstration of this phenomenon is the Josephson effect in
superconductors, which was predicted theoretically by B. Josephson (then a graduate student!) in 1962
and observed experimentally in less than a year. (More recently, this effect was also observed in atomic
Bose-Einstein condensates.) Experimentally, the easiest way to observe the effect is two connect two
superconductor samples with some sort of weak electric connection (“weak link”) and bias them a
constant (dc) voltage V, typically in a few-microvolt range – see Fig. 3.
I  sin(1   2 )
 expi1   expi 2 
9 A good argument here is that in nonrelativistic mechanics the potential energy is defined only to an arbitrary
constant. V Fig. 2.3. Josephson effect in superconductivity.
Quantum Mechanics
The theory of superconductivity shows that it may be explained by the electron merging, at low
temperatures, into the so-called Cooper pairs of electrons (each consisting of two electrons with
opposite spins and momenta), which form a coherent Bose-Einstein condensate – see, e.g., SM Sec. 3.5.
Most properties of such a condensate may be described by a single wavefunction, evolving in time in the
effective potential energy U = -2e + const, where  is the electrochemical potential, and (-2e) is the
total charge of the Cooper pair. As a result, for the situation shown in Fig. 3, Eq. (27) takes the form
2e
 Vt  const, (2.28)

where V is the difference of electrochemical potentials, i.e. the applied voltage. B. Josephson has shown
that current I of Cooper pairs through a weak link should be a 2-periodic function of this difference,10
which in many cases may be approximated by a sine:
I  I c sin  , (2.29)
where Ic is some constant (depending , in particular of the weak link strength). Combining Eqs. (28) and
(29), we see that the current oscillates in time with the so-called Josephson frequency
1 2e 2e
fJ  V  V (2.30)
2  h
as high as ~ 484 MHz per each microvolt of applied voltage. This effect is not only well documented,
but also is being used in highly accurate primary standards of dc voltage.11
Now, let us discuss the particle motion in a potential U(x) which is not constant (Fig. 4). In
classical mechanics, the full mechanical energy
p2
E  U ( x) (2.31)
2m
of a particle moving in the field of conservative forces, is conserved – see the horizontal line in Fig. 4.
10 See, e.g., EM Sec. 6.3 for simple arguments leading to Eq. (22).
11 The most precise proof that the Josephson frequency-to-voltage ratio fJ/V does not depend on superconducting
material (to 15 decimal places!) was carried out by Prof. J. Lukens’ group here at Stony Brook – see J.-S. Tsai et
al., Phys. Rev. Lett. 51, 316 (1983).
Quantum Mechanics
classically accessible classically forbidden
U (x) Fig. 2.4. Classical 1D motion in a potential

xc profile U(x).
classical turning point
If E > U(x) within the whole region of our interest, kinetic energy p2/2m never turns to zero. This means
that the particle continues to move in the initial direction. On the other hand, if U(x) becomes larger than
E (Fig. 4), the particle cannot penetrate that “classically forbidden region” (because its kinetic energy
cannot be negative), and is reflected from the “classical turning point” xc defined by equation
U ( xc )  E . (2.32)
In order to see what wave mechanics says about this situation, consider, as the simplest case, the sharp
“potential step” shown in Fig. 5:
 0, at x  0,
U ( x)   (2.33)
U 0 , at 0  x.
U ( x), E
A C
 ( x) U0
B Fig. 2.5. Reflection of a
monochromatic wave from a
potential step U0 > E. (This particular
wavefunction shape is for U0 = 5E.)
E
The wavefunction is plotted with the
same schematic vertical offset by E,
0 x
as those in Fig. 1.7.
We consider a fixed the particle energy (i.e., speak about the limit of a very long wave packet),
so that the time dependence of the solution is already known – see Eq. (1.45). This we can limit
ourselves to the solution of the 1D version of the stationary Schrödinger equation (1.48):
 2 d 2
  U ( x)  E (2.34)
2m dx 2
for the spatial part (x) of the wavefunction. Notice that this is not exactly the eigenproblem like the
one we have solved in Sec. 1.4 for a quantum well, in the sense that now energy E is considered fixed –
e.g., by the initial conditions which launch a long wave packet on the potential step, from the left.
At x < 0, i.e. at U = 0, the equation may be satisfied with either of two traveling waves,
proportional to exp{+ikx} or exp {-ikx}, with both values of wavevector satisfying equation (2):
Quantum Mechanics
2mE
k2  . (2.35)
2
In the past section we could discuss these solutions separately, but due to the physics of the problem we
are discussing now (particle reflection), we should take the general solution
 ( x  0)  A exp{ikx}  B exp{ikx} . (2.36)
(Actually, we have already used such solution, with B = -A, when analyzing the potential well – see Eq.
(1.55), but now the relation between coefficients B and A may be different, because of the finite height
U0 of the step.)
At x > 0, i.e. U = U0 > E, Eq. (34) may be rewritten as
d 2
  2 , (2.37)
dx 2
with
2m(U 0  E )
2   0. (2.38)
2
The general solution of Eq. (37) is the sum of exp{+x} and exp{-x}, with arbitrary coefficients.
However, in physics the wavefunction should be finite at x  , so only the latter exponent is
acceptable:
 ( x  0)  C exp{x} . (2.39)
This exponential decay of the wavefunction in the classically forbidden region, and hence a finite
probability to find the particle there, is one of the most fascinating predictions of quantum mechanics,
and has been repeatedly observed in experiment, e.g., via tunneling experiments – see below. It is
evident that constant , defined by the second of Eqs. (43), may be interpreted as the reciprocal
penetration depth of the wavefunction into the classically forbidden region. Even for the lightest
particles this depth is extremely small. Indeed, for E << U0 that equation yields
1 
 E 0  . (2.40)
 2mU 0
For example, for a conduction electron in a typical metal, which runs, at its surface, into a sharp
potential step U0 whose height equals to metal’s workfunction W  5 eV (see the discussion of the
photoelectric effect in Sec. 1.1),  is close to 0.1 nm, i.e. 1 angstrom. For heavier elementary particles
(e.g., a proton) the penetration depth is correspondingly lower, and for a macroscopic bodies is hardly
measurable.
Coefficients A, B, and C should be found from the boundary conditions at x = 0. Since E is a
finite constant, and U(x) is finite, Eq. (34) says that d2/dx2 should be finite as well. This means that the
first derivative should be continuous:
 
 d d  d 2 2m
lim  0  x   x     lim  0  2 dx  2 lim  0  U ( x)  E  dx  0 . (2.41)
 dx dx   dx  
Quantum Mechanics
Repeating the calculation for function (x) itself, we see that it also should be continuous at all point,
including x = 0. Plugging solutions (36) and (39) into these two boundary conditions, we get a system of
two linear equations
A  B  C , ikA  ikB  C , (2.42)
whose (elementary) solution enables us to express B and C via A :
k  i 2k
BA , CA . (2.43)
k  i k  i
We immediately see that since the nominator and denominator in the first of these formula have
equal moduli, i.e. as we could expect, the particle is completely reflected from the step. As a result, at x
< 0 our solution (36) may be presented by a standing wave
k
 ( x  0)  2iAe i sin( kx   ),   arctan . (2.44)

Notice that the shift x  /k of the standing wave to the right is commensurate with, but not equal to 
 1/. (At E  0, x  /2.) Figure 5 shows the full behavior of the wavefunction, for a particular case
E = U0/5, i.e. k/ = [E/(U0-E)]1/2= 1/2.
According to Eq. (38), as the particle’s energy E is increased to approach U0, the penetration
depth diverges. This raises an important issue: what happens at E > U0, i.e. if there is no classically
forbidden region? Again, in classical mechanics the incident particle would continue to move to the
right, though with a reduced velocity, corresponding to the new kinetic energy E – U0. In quantum
mechanics, however, the situation is different. In order to analyze it, it is not necessary to redo the whole
problem; it is sufficient to notice that all our calculations, and hence Eqs. (43) are still valid if we take
2m( E  U 0 )
  ik ' , with k '2   0. (2.45)

With this change, Eq. (43) yields
k  k' 2k
BA , CA . (2.46)
k  k' k  k'
The most important result of this change is that now the reflection is not complete: B  A. In order to
evaluate this effect qualitatively, it is more fair to use not the B/A or C/A ratios, but rather that of the
probability currents (1.?) corresponding to traveling waves with amplitudes C and A, in the
corresponding regions (respectively, x > 0 and x < 0):
2
J k' C 4kk ' 4 E(E  U 0 )
T C    . (2.47)
JA kA
2
(k  k ' ) 2  E  E U0 
2
(T is called the transparency of the inhomogeneity, in our current case the potential step.) The result
given by Eq. (47) is plotted in Fig. 6a. Notice the most important features of this result:
(i) At E  U0, the transparency tends to zero, giving a proper connection with the case E < U0.
Quantum Mechanics
(ii) We can use result (47) even for U0 < 0, i.e. for the “step down” (or “cliff”) profile – see Fig.
6b. Very counter-intuitively, the particle is (partly) reflected even from such a cliff, and the transmission
diminishes (rather slowly) as U0  -.
(a) (b)
1 A C
E0
0.8 B0
U 0
0.6
T U0
0.4
0.2 Fig. 2.6. (a) Transmission coefficient of a potential

step as a function of its height, according to Eq. (), and
0 (b) a scheme of wave components at U0 < E (shown
1 0 1
for U0 < 0).
U0 / E
The most important conceptual conclusion of our analysis is that the quantum particle is “partly
reflected” from a potential step with U0 < E, in the sense that there is a nonvanishing probability T < 1
to find it passed over the step, while there is also probability (1 – T) to have it reflected.
The same property is exhibited, for any relation between E and U0, by another simple system, the
famous tunnel barrier profile. Figure 7 shows its simplest, “rectangular” version:
 0, for x  d / 2,

U ( x)  U 0 , for  d / 2  x   d / 2, (2.48)
 0, for  d / 2  x.

U  U0
A C F
E
B D
U 0 Fig. 2.7. Rectangular tunnel barrier.

d /2 d /2 x
In order to analyze this problem, it is sufficient to look for the solution to the Schrödinger
equation in the form (36) at x  -d/2. At x > +d/2, i.e., behind the barrier, we may keep just one
traveling wave,
 ( x)  F exp{ikx} , (2.49)
because there is no inhomogeneity beyond point +d/2, which could reflect the wave back. However,
under the barrier, i.e. at -d/2  x  +d/2, we should generally keep both exponential solutions,
 ( x)  C exp{x}  D exp{x} , (2.50)
Quantum Mechanics
because our previous argument, used in the potential step problem, is no longer valid. (Here k and  are
still defined, respectively, by Eqs. (35) and (38).) In order to find the relation between coefficients A, B,
C, D, and F, we need to plug in the solutions into the same conditions of continuity of the wavefunction
and its first derivative, now at two boundary points, x =  d/2. Solving the resulting system of 4 linear
equations for five amplitudes (A, B, C, D, and F), we can calculate all ratios B/A, C/A, etc., in particular
the ratio
F exp ikd 
 , (2.51)
A i  2  k2 
cosh d    sinh d
2  k 
whose modulus squared gives the barrier transparency T. (Notice that factor k’/k, which participates in
Eq. (47), equals 1 in our current problem, because the potential energy is the same before and after the
barrier.) Figure 8a shows the transparency as a function of particle energy E, for several characteristic
values of the barrier thickness d, or rather ratio d/, where  is defined by Eq. (44).
(a) (b)
0.01
d 6
d /   3 .0
0.8  0.3 110

2  10
110 10
0.6
 14
110
T 1 .0 T
 18
0.4 110 30
 22
110
0.2 3 .0
 26
110
 30
110
0 1 2 3 0 0.2 0.4 0.6 0.8
E /U 0 1  E / U 0  1/ 2
Fig. 2.8. Transparency of the rectangular tunnel barrier as a function of particle’s energy E.
The plots show that for a thin barrier (d < ) the transparency grows gradually with particle’s
energy. This growth is natural, because the penetration constant  decreases with the growth of E, i.e.,
the wavefunction penetrates more and more into the barrier, so that more and more of it is “picked up”
at the second interface (x = +d/2) and transferred into the wave Fexp{ikx} propagating beyond the
barrier. For thick barriers (d >> ) , this dependence is dominated by an exponent,
2
 4k 
T  2 2 
exp{2d } , (2.52)
 k  
and may be well seen as a straight segments in semi-log plots (Fig. 8b) of T as a function of (U0 – E)1/2
which is proportional to  - see Eq. (38).
Equation (52) also clearly shows the exponential dependence of the barrier transparency of its
thickness d >> . This dependence is the most important factor for various applications for quantum-
Quantum Mechanics
mechanical tunneling – from the field emission12 of electrons to scanning tunneling microscopy.13 (This
effect also has huge negative implications for modern electronic engineering, for example, limiting the
further scaling down of field effect transistors in semiconductor integrated circuits, due to increasing
tunneling through the gate oxide.)
Another interesting effect visible in Fig. 8a for the case a = 0.3, are the oscillations of T at E >
U0. This is our first example of resonant tunneling - see Exercise 2 and Sec. 4 below.
2.3. The WKB approximation

Before moving on to exploring more complex potentials, let us see whether the results discussed
in the previous section hold on in the opposite limit of “soft”, gradual potential profiles. (A quantitative
condition of the “softness” will be derived below). The most efficient analytical tool in this limit is the
“quasiclassical” (or “WKB”) approximation developed by H. Jeffrey, G. Wentzel, A. Kramers, L.
Brillouin in 1926-27.
In order to derive its 1D version, let us rewrite Schrödinger equation (34) as
 " k 2 ( x)  0 (2.53)
where prime is used to denote differentiation over x, and the wave vector is defined just as in Eq. (45),
2mE  U ( x)
k 2 ( x)  , (2.54)
2
though now it may be a function of x. We already know that for k(x) = const, the fundamental particular
solutions to this equation have form (x) = A exp{ikx}. Any of them may be presented in a simple form
 ( x)  exp{i ( x)} , (2.55)
where (x) is a complex function, in this simplest case equal to (x) = kx – ilnA. This is why we may
try to use the same Eq. (55) as a particular solution of Eq. (53) in the general case, k(x)  const, but with
arbitrary (x). Calculating the second derivative of this wavefunction,
 '  i' exp{i} ,  "  i"( ' ) 2 exp{i} , (2.56)
Plugging it into Eq. (53), and requiring that the net factor before exp{i} is zero, we get
i"( ' ) 2  k 2 ( x)  0 . (2.57)
This is still an exact, general result. At first sight, it looks worse than the initial equation (53), because
Eq. (57) is nonlinear. However, it is more ready for simplification in the limit U’  0. Indeed, we know
that for a uniform potential, ” = 0. Hence, in the “0th” approximation we may try to keep that result
and take (x) = 0(x), with
( '0 ) 2  k 2 ( x) . (2.58)
Just as in the uniform case, this equation has two roots,
12 See, e.g., G. N. Fursey, Field Emission in Vacuum Microelectronics, Kluwer, New York, 2005.
13 See, e.g., G. Binning and H. Rohrer, Helv. Phys. Acta 55, 726 (1982).
Quantum Mechanics
 '0   k ( x) , (2.59)
(where we will take positive root for k(x)), so that the general solution is
 x   x 
 0 ( x)  A exp i  k ( x' )dx'  B exp i  k ( x' )dx' . (2.60)
   
The physical sense of this result is simple: it is a sum of forward- and back-propagating waves, with the
coordinate-dependent local vector number (54) which self-adjusts to the potential profile.
Let me emphasize the non-trivial nature of this approximation.14 First, any attempt to address the
problem with the standard perturbation approach (say,  = 0 + 1 +… ) would fail for most potentials,
because even a slight but persisting deviation of U(x) from a constant level leads to a gradual
accumulation of phase 0, impossible to describe by any small perturbation of .
Second, the dropping of term ” is hard to justify. Indeed, since we are committed to the “soft
potential” limit U’  0, we should be ready to assume that the characteristic length a of variation of 
has to be large, and neglect the terms which are the smallest ones in the limit a  . However, both
first terms in Eq. (57) are apparently of the same order in a, O(a-2), why have we neglected just one of
them?
The price we have paid for that questionable treatment is high: Eq. (60) does not satisfy the
fundamental property of the Schrödinger equation, the probability current conservation. Indeed, since
Eq. (34) describes a fixed-energy (stationary) spatial part of the general Schrödinger equation, its
probability density w = * =*, and should not depend on time. Hence, according to Eq. (1.?), we
should have J(x) = const. However, this is not true for each component of Eq. (6); for example for the
forward-propagating wave, Eq. (1.?) yields
 2
J 0 ( x)  A k ( x) , (2.61)
m
evidently not a constant if k(x)  const.
The brilliance of the WKB theory is that the both problems may be fixed without revising the 0th
approximation. Indeed, let us explore the next (1st) approximation instead,
 ( x)   0 ( x)   1 ( x) , (2.62)
where 0 still follows Eq. (59), while 1 describes a small correction:

 1'   '0  k ( x) . (2.63)
Plugging Eq. (62) into Eq. (57), we get

 
i  "0   1"   1' (2 '0   1' )  0 . (2.64)
Using condition (63), we can neglect 1” in comparison with 0” in the first parenthesis, and 1’ in
comparison with 20’ in the second parenthesis. As a result, we get an approximate result
14 Philosophically, it is very close to the “small parameter” methods in the theory of weakly nonlinear
oscillations and waves – see, e.g., CM Ch. 4.
Quantum Mechanics
 
i  "0 i
'
1
2  '0 2

 ln k ( x)'  i ln k ( x) ' ,  (2.65)
so that
x
1
i WKB  i 0  i 1  i  k ( x' )dx'  ln , (2.66)
k ( x)
and, finally,
a  x  b  x 
 WKB ( x )  expi  k ( x' )dx'  exp i  k ( x' )dx', for k 2  0. (2.67)
k ( x)   k ( x)  
(The lower integration limit is arbitrary, but its choice evidently affect constants a and b.)
This modification of the 0th approximation (60) overcomes the problem of current continuity; for
example, for the forward-propagating wave
 2
J WKB ( x)  a  const . (2.68)
m
Physically, the factor k in the denominator of the WKB wave is easy to understand. When U(x)
increases and approaches E, kinetic energy of the particle, p2/2m = E – U(x), drops, and so does its
momentum and speed. Hence it should be “easier” (more probable) to find the particle within a certain
interval dx. This is exactly the result which WKB gives: dP/dx = w(x) = *  1/k(x) = /p(x).
Another achievement of the first approximation is a clear understanding of the validity of the
WKB theory: it is given by Eq. (63). Plugging into it our result (65), and estimating it as ~ ln(ka)/2a, we
can rewrite the validity condition as
ln(ka)  2ka . (2.69)
Since logarithm is, for large values of its argument, a very slow function, in most situations this
condition may be presented as just ka >> 1, reading that the region where U(x) changes substantially
should contain many de Broglie wavelengths.
Now let us extend the WKB approximation to the situation where the difference U(x) – E may
change sign, for example to the reflection problem sketched in Fig. 6. Just as we did for the sharp
potential step, we first need to find the solution for the classically forbidden region, in this case x > xc.
(Equation (67) is completely adequate for the region x < xc.) For that, there is no need to redo our
calculations, because they are still valid if we, just as in the sharp step problem, take k(x) = i(x), where
2mU ( x)  E 
 2 ( x)   0, for x  xc . (2.70)
2
and take into account just one of two possible solutions (with  >0), in analogy with Eq. (39). The result
is
c  x 
 WKB ( x)  exp   ( x' )dx', for k 2  0, i.e. κ 2  0. (2.71)
 ( x)  
Quantum Mechanics
This is a really great formula! It described the quantum-mechanical penetration of the particle into the
classically forbidden region, and provides a natural generalization of Eq. (39) for the constant potential
(leaving intact, of course, our approximate estimates of the depth  ~ 1/ of such penetration).
At this stage, acting as in the sharp-step problem in Sec. 2, we would use the boundary
conditions in the interface point x = xc to relate the wave “amplitudes” a, b, and c. However, now this
operation is a tad more complex, because both WKB functions (67) and (71) diverge, albeit weakly, at
the classical turning point, were both k(x) and (x) tend to zero. This “connection problem” may be
however, solved in a different way. Let us use the commitment of potential “softness”, assuming that it
allows us to use just two leading terms in the Taylor expansion of function U(x) near point xc:
dU
U ( x)  U ( xc )  x  xc ( x  xc )  E  U ' ( x  xc ) . (2.72)
dx
After introducing a dimensionless variable
1/ 3
x  xc  2 
  , x0    , (2.73)
x0  2 mU ' 
this approximation reduces the Schrödinger equation (53) to the so-called Airy equation
d 2
   0 . (2.74)
d 2
As all linear, ordinary differential equations of the second order, the general solution to this equation
may be presented as a linear combination of two fundamental solutions, in this case called Airy
functions Ai( ) and Bi( ) shown in Fig. 9a.15 The latter function diverges at   , and thus is not
suitable for our current problem (Fig. 4), while the former function has the following asymptotic
behavior at   >> 1:
 1  2 3/ 2 
1  2 exp 3  , at   ,
Ai( )     (2.75)
 1/ 2 
1/ 4
 2 
sin    3 / 2  , at   .
  3 4
Now let us apply to the Airy equation the WKB approximation (using dimensionless variables). For
that, we first need to calculate the WKB exponents. Taking the classical turning point ( = 0) for the
lower limit. Then, in dimensionless units, for  > 0 we get

2
 2 ( )   ,  ( )   1 / 2 ,   ( )d
0
  3/ 2 ,
3
(2.76)
15 The following integral formulas,

1   3 
 
1  3    3
Ai( )   cos
 0  3
   d ,

Bi( )   exp
 
 0  3
    sin     d ,
  3 
are more convenient for practical calculations of the Airy functions than the differential equation (74).
Quantum Mechanics
i.e. exactly the exponent in the first line of Eq. (75). Making a similar calculation for  < 0, with the
natural assumption b = a (full reflection from the potential step), we finally arrive at
  2 
 c exp  3 / 2 , at   0,
1   3 
Ai WKB ( )  1 / 4   (2.77)
 a sin  2   3 / 2   , at   0.
 3 
We see that though this approximation differs from the exact solution at small values of  (Fig. 9b), i.e.
close to the classical turning point, their asymptotic behaviors coincide if16
 a
 and c  . (2.78)
4 2
The first result may be rewritten in a different form:
a'   x   x  
 WKB ( x  x c )  exp i  k ( x' )dx'  exp i  k ( x' )dx'  i  . (2.79)
k ( x)   xc   xc 2 

. (a) (b)
1 1
Bi ( ) Ai WKB ( )
Ai( ) Ai( )
0 0
)
1 1
 10 0 10 3 0 3
 
Fig. 2.9. (a) Airy functions Ai and Bi, and (b) the WKB approximation for function Ai.
It gives a good mnemonic rule: reflecting from a “soft” potential step, the wavefunction acquires an
additional phase shift  = /2, if compared with the reflection from the “vertical” potential wall
Let us compare this result with that for a sharp but finite potential step with U0 > E. Using our
result (43)-(44), we can rewrite Eq. (36) in the form similar to Eqs. (79)
  x
  x
 k
 ( x  0)  A' expi  k ( x)dx   exp i  k ( x)dx  i ,   2  2 arctan , (2.80)
  0   0  
we see that the additional phase shift depends on particle’s energy:
16 An alternative way to derive these “connection formulas” (not involving the Airy functions but using requiring
the analytical extension of WKB formulas to the plane of complex argument x) may be found in Sec. 47 of the
textbook by Landau and Lifshitz.
Quantum Mechanics
E
  2 arctan , (2.81)
U0  E
and may be anywhere between 0 (for E << U0, e.g., an effectively infinite wall) and  (for E  U0), and
coincides with the WKB value only for a particular value E = U0/2.
By the way, the virtual identity of the two solutions, (75) and (77), at   > 1 immediately
enables us to formulate the condition of validity of the WKB approximation for the step problem - in
other words, the criterion of the step “softness”.17 In order to have the connection formulas valid for
some potential profile U(x), the region of the WKB approximation validity for the Airy equation
(formally,   >> 1, i.e. x - xc >> x0) should overlap with the region where the linear approximation
(2.72) is valid. The width a of the latter region is of the order of U’/U” (with both derivatives evaluated
at xc), so that the overlap region is wide if
1/ 3 4
U'  2  3 2m U '
a~  x 0    , i.e. if U "  , (2.82)
U"  2mU '  2
where in the last form we have replaced U’ for U’ to make the softness condition applicable to potential
steps with negative slope (whose analysis is absolutely similar) as well.
Let us apply the WKB approximation to calculate the energy spectrum of 1D particle in a
quantum well with “soft” walls (Fig. 10). We can always consider the standing wave describing an
eigenstate n (corresponding to eigenenergy En) as a traveling wave going back and force between the
walls, being sequentially reflected by each of them. Let us apply the WKB approximation to such a
traveling wave. First, propagating from the left classical turning point xL to the right point xR, it acquires
phase change
xR
    k ( x)dx . (2.83)
xL
At the reflection from the soft wall at xR, according to the connection formula, the wave acquires an
additional shift /2.18 Now, traveling back from xR to xL the wave gets a shift similar to one given by
Eq. (83):  = . Finally, at the reflection from xL it gets one more /2. Summing up all these
contributions, we may write the self-consistency condition (that the wavefunction “catches its own
tail”), in the form
xR
 
 total  2      2  k ( x)dx    2n, with n = 1, 2, … (2.84)
 2 xL
In terms of particles momentum p(x) = k(x), this gives the famous Bohr-Sommerfeld quantization rule
17 Condition (69) is somewhat ambiguous for this case, because due to the divergence of k at the classical turning
point, it is not immediately clear which value of k can be used, and what exactly is a.
18 The WKB approximation is also applicable to a quantum well with a softly bending “floor”, but sharply rising
walls – see, e.g., Homework Problems 3.1. In this case we may still use Eq. (83), but the phase shift at the
reflection from the vertical wall should be taken from Eq. (81).
Quantum Mechanics
1 1
 p( x)dx  2(n  2 )  h(n  2 ) ,
C
(2.85)
where the closed path C means the full period of classical motion.19
U ( x)
En
Fig. 2.10. Quasiclassical treatment of eigenstates in a

1D potential well.
xL 0 xR x
Let us see what this rule gives for the very important case of a quadratic potential profile of a
harmonic oscillator of frequency 0. In this case,
m 2 2
U ( x)  0 x , (2.86)
2
and the classical turning points are the roots of a simple equation
m 2 2
 0 xc  E n , (2.87)
2
so that xR = xc > 0, xL = - xc < 0. Due to potential’s symmetry, the required integration is also simple:
xR xc xc xc
2m[ En  U ( x)] 2mEn x2 E
 k ( x)dx  2  k ( x)dx  2 
xL 0 0
2
dx  2
 
0
1  2 dx   n ,
xc 0
(2.88)
so that Eq. (84) is satisfied if

 1
E n   0  n' , with n’  n - 1 = 0, 1, 2, … (2.89)
 2
In order to estimate the validity of this result, we have to check condition (69) in a typical point
of the classically allowed region (say, x = 0), and Eq. (79) at the turning points. A straightforward
calculation shows that both conditions are valid for n >> 1. However, we will see below that Eq. (89) is
actually exactly correct for all energy levels, due to special properties of potential (86).
Now, let us look at the second connection formula, c = a/2. Again, it differs from the result (43)
for a sharp potential step, which may be rewritten as
2k 2
CA A exp i 2  , (2.90)
k  i 1  ( / k ) 2
19 Notice that at motion in more than one dimension, a closed classical trajectory may have no turning points. In
this case, ½ in the parentheses of Eq. (85), arising from the turns, should be dropped. The simplest example is the
circular motion of the electron about the proton in Bohr’s picture of the hydrogen atom, for which the modified
quantization condition takes form (1.10).
Quantum Mechanics
by both the modulus and the phase factor. (In the WKB approximation, the latter always equals /4.)
Hence, predictions of the WKB approximation are not valid for arbitrary potentials. Nevertheless, these
predictions are an important part of practical applications of wave mechanics. One of the most important
of them is the transparency of an arbitrary potential barrier (Fig. 11).
U ( x) d  xc'  xc
U max
a c f
E
b d
Fig. 2.11. 1D potential barrier of
0 an arbitrary shape.
xc x c' x
Here, just as in the case of a rectangular barrier, we need to take unto consideration five
particular “waves” (or rather fundamental solutions): 20
 a x  b  x 
 expi  k ( x' )dx'  exp i  k ( x' )dx', for x  xc ,
 k ( x) 0  k ( x)  0 
 c  x  d x 
 WKB  exp   ( x' )dx'  exp  ( x' )dx', for xc  x  xc' , (2.91)
  ( x)  0   ( x) 0 
 f  x

 expi  k ( x' )dx', for x c'  x.
 k ( x) 0 
Since on the right of the left classical point we have two exponents rather than one, and on the
right of the second point, one traveling waves rather than two, the connection formulas (78) have to be
generalized, using asymptotic formulas not only for Ai( ), but also for the second Airy function, Bi( ).
The analysis (very straightforward but a bit bulky) may be found, for example, in Sec. 7.4 of
Merzbacher’s textbook. The final result of that calculation is remarkably simple:
 xc   2 xc 
' '
2
f
T  exp 2   ( x)dx   exp  2mU ( x)  E dx  . (2.92)
a  xc    xc 
with no pre-exponential factor. This formula is broadly used in applied quantum mechanics, despite the
approximate character of its pre-exponential coefficient for insufficiently soft barriers. (For example,
Eq. (52) shows that for a thick rectangular barrier with k = , i.e. U0 = 2E, the WKB approximation (92)
underestimates T by a factor of 4. However, on the logarithmic scale of Fig. 8b, such factor still looks as
a small correction.)
Notice that in order to have the WKB validity conditions (69) and (82) satisfied, the transparency
T of a smooth barrier must not necessarily be much lower than 1. Hence we can use it to examine the
limit when E approaches the barrier top Umax (Fig. 11). Here points xc and xc’ merge, so that, according
20 Sorry, but the same letter, d, is used here for the barrier thickness (defined in this case as the classically
forbidden region length, xc’ – xc), and the constant in one of the wave amplitudes – see Eq. (90). Let me hope that
the difference between these uses is absolutely evident from the context.
Quantum Mechanics
to Eq. (92), T  1. Thus, we see a major difference between smooth and sharp barriers: in the former
case, there is no partial reflection at passage over the barrier! This fact is also evident from the direct
WKB analysis of the case E > Umax, because an adequate solution is given by just one, quasiclassical
traveling wave
a  x 
 WKB ( x )  expi  k ( x' )dx' , (2.93)
k ( x)  
which conserves the probability current.

For arbitrary potentials there is always some partial reflection at the over-barrier motion (see,
e.g., the Kemble formula cited in Homework 3), but its intensity drops rapidly as soon as the barrier
edges are smoothed.
Our discussions of the propagator and WKB approximation have opened a straight way toward
the Feynman path integral, but I will postpone this step until a more compact bra-ket notation has been
introduced in Chapter 4.
2.4. More 1D: transfer matrix, resonant tunneling, and metastable state lifetime
Let us now explore motion in more complex potentials. The piecewise-constant and smooth-
potential models are not very good tools here, because they both require “stitching” local solutions in
each classical turning point, which may lead to very cumbersome calculations. However, we may get a
very good insight of physics using potentials composed of several -functions.
Let us start with looking at what our result (51) gives in the limit of a very thin and high
rectangular barrier, d << 1/, U0 >> E:
2
F 1 1
T   , (2.94)
A 1  i
2
1  2
where
1  2  k2  1  2d m
   d   2 U 0d . (2.95)
2  k  2 k  k
The last product, U0d, is just the “area”
W  U ( x)dx
U ( x ) E
(2.96)
of the barrier, and this fact implies that the very simple result (93) for the transparency may be correct
for a barrier of any shape, provided that it is sufficiently thin and high.
Indeed, let us consider the tunneling problem for a very thin barrier with d, kd << 1 (Fig. 12),
presenting it by a -function:
U ( x)  W ( x) . (2.97)
Quantum Mechanics
We already know the solutions in all points but x = 0 – see Eqs. (36) and (49) – so we need just to
analyze boundary conditions in that point to find coefficients A, B, and F (or rather their ratios).
However, due to the special character of the -function, we should be careful here. Indeed, Eq. (41) is
still valid, but integral in the right-hand part is now finite for any , giving

d d 2m 2m
2 
x 0  x0  lim  0 U ( x) dx  2 W (0). (2.98)
dx dx   
U ( x )  W ( x )
A F
E
B
Fig. 2.12. -functional tunnel barrier.
0 x
On the other hand, the wavefunction itself is still continuous:


d
 x 0  x0  lim  0

 dx dx  0. (2.99)
Using these boundary conditions, we readily get the following system of two linear equations,
2mW
A  B  F , ikF  (ikA  ikB)  F, (2.100)
2
which yields
B  i F 1 mW
 ,  ,  2 . (2.101)
A 1  i A 1  i  k
We see that we have again arrived at Eq. (94).
We can use this result to derive the simple expression for the energy dependence of the
transmission coefficient:
1 E mW 2
T  , E0  , (2.102)
1   2 E  E0 2 2
(valid only for E << Umax) which shows that as energy becomes larger than E0, the barrier’s transparency
approaches unity. However, the most important application of Eqs. (100) is for deriving transparency of
more complex potential profiles. For that, let us first introduce very general notions of the 1D scattering
matrix and transfer matrices.
Consider an arbitrary but finite-length potential “bump” (more formally called a “scatterer”),
extending from x1 to x2, on the flat potential background, say U = 0 (Fig. 13). We know the general
solution, with a certain energy E, outside the interval are simple waves. Let us present them in the form
 k  Ak exp ik ( x  x k )  Bk exp ik ( x  x k ), (2.103)
Quantum Mechanics
where (for now) k = 1 or 2, and (k)2/2m = E. Notice that each wave has, in this notation, its own
reference point, because this is very convenient for what follows.
U ( x)
A1 A2
E
B1 B2
Fig. 2.13. A single 1D scatterer.

x1 0 x2 x
As we have already discussed, if the wave/particle is incident from the left, the Schrödinger
equation within the scatterer range (x1 < x < x2), can provide only linear expressions of the transmitted
(A2) and reflected (B1) wave amplitudes via the incident wave amplitude A1:
A2  S 21 A1 , B1  S11 A1 , (2.104)
where S11 and S21 are certain (generally, complex) numbers – for specific examples we have already
examined see, e.g., Eqs. (43), or (51), or (101). In this case, B2 = 0. Alternatively, if a wave, with
amplitude B2, is incident from the right, it also may induce a transmitted wave (B1) and reflected wave
(A2) with amplitudes
B1  S12 B2 , A2  S 22 B2 , (2.105)
where coefficients S22 and S12 are generally different from S11 and S21. Now we can use the linear
superposition principle to argue that if waves A1 and B2 are simultaneously incident on the scatterer
(say, because wave B2 has been partly reflected back by some other scatterer located at x > x2), the
resulting “scattered wave” amplitudes A2 and B1 are the sums of what we have had for separate incident
waves:
B1  S11 A1  S12 B2 ,
(2.106)
A2  S 21 A1  S 22 B2 .
These linear relations may be conveniently presented by the so-called scattering matrix S:
 B1  A  S S12 
   S  1 , S   11 . (2.107)
 A2   B2   S 21 S 22 
Scattering matrices, duly generalized, are very important notions for wave scattering in more
than one dimensions; for 1D problems, however, another matrix is more convenient to present the same
linear relations (106). Let us solve this system of linear equations for A2 and B2:
A2  T11 A1  T12 B1 , A  A 
i.e.  2   T  1 , (2.108)
B2  T21 A1  T22 B1 ,  B2   B1 
where T is the transfer matrix with elements
S11 S 22 S S 1
T11  S 21  , T12  22 , T21   11 , T22  . (2.109)
S12 S12 S 21 S12
Quantum Mechanics
Actually, due to symmetry properties of the transfer matrix (see below), T11 may be presented in a
simpler form, similar to T22: T11 = 1/S21*. This relation allows a ready expression of scatterer’s
transparency via just one coefficient of the transfer matrix:
2
A 2 2
T 2  S 21  T11 . (2.110)
A1 B2  0
The most important property of 1D transfer matrices is that in order to find the total transfer
matrix T of a system consisting of several, say N, sequential but arbitrary scatterers (Fig. 14), it is
sufficient to multiply their matrices. Indeed, extending definition (108) to other points xj (j = 1,…N+1),
we can write
A1 A2 A3 AN 1
B1 B2 B3 BN 1
Fig. 2.14. A sequence of several 1D
 scatterers.
x1 x2 x3 x N 1 x
 A2  A   A3  A  A 
   T1  1 ,    T2  2   T2 T1  1 , (2.111)
 B2   B1   B3   B2   B1 
etc. (where the matrix indices indicate the scatterers’ order), so that
 AN 1  A 
   TN TN 1 ...T1  1 . (2.112)
 B N 1   B1 
But we can also define the total transfer matrix as
 AN 1  A 
   T 1 , (2.113)
 B N 1   B1 
so that finally
T  TN TN 1 ...T1 . (2.114)
It is possible to show that this formula is valid even if the flat-potential gaps between component
scatterers vanish, so that it may be applied to a scatterer with an arbitrary profile U(x), by fragmenting
its length into small segments x = xj+1 - xj, and treating each fragment as a rectangular barrier of height
(Uj)ef = [U(xj+1) – U(xj)]/2 - see Fig. 15. Since very efficient numerical algorithms are readily available
for fast multiplication of matrices (especially as small as 2×2), this approach is broadly used in practice
for computation of transparency of tunnel barriers of complex form. (It is much more efficient then the
direct numerical solution of the Schrödinger equation.)
Quantum Mechanics
U ( x)
A1 (U j ) ef AN 1
B1 BN 1
Fig. 2.15. The transfer matrix approach
to a long tunnel barrier of an arbitrary
profile.
x1 x2    x j x j 1    x N 1 x
In order to use this approach for several conceptually important systems, let us calculate the
transfer matrices for elementary scatterers, starting from the -functional barrier located at x = 0 – see
Eq. (97). Taking x1 = x2 = 0, we can merely change the notation in Eq. (100) to get
 i 1
S11  , S 21  . (2.115)
1  i 1  i
Repeating the analysis for the wave incident from the left, we get
 i 1
S 22  , S12  , (2.116)
1  i 1  i
and using Eqs. (109), we get
 1  i  i 
T   . (2.117)
 i 1  i 
The next example may seem strange at the first glance: what if there is no scatterer at all between
points x1 and x2? If points x1 and x2 coincide, the answer is indeed trivial and can be obtained, e.g., from
Eq. (110) by taking W = 0, i.e.  = 0:
1 0
T0     I (2.118)
0 1
- the so-called identity matrix. However, we are free to choose the reference point x1,2 participating in
Eq. (103) as we wish. For example, what if x2 – x1 = a  0? Let us first take the forward-propagating
wave alone: B2 = 0 (and hence B1 = 0); then
 2   1  A1 expik ( x  x1 )  A1 expik ( x 2  x1 )expik ( x  x 2 ), (2.119)
which means A2 = A1 exp{ik(x2 - x1)} = A1 exp{ika}, i.e. T11 = exp{ika}. Repeating the calculation for
the back-propagating wave, we see that T22 = exp{-ika}, and since there is no reflection
(“backscattering”), we finally get
 e ika 0 
Ta   , (2.120)
 0 e 
ika
Quantum Mechanics
independently of the mutual position of points x1 and x2. (At a = 0, we naturally recover the special case
(118).)21
Let us use these results to calculate the transparency T of the double-barrier system shown in
Fig. 14. We could of course calculate it as before, writing down explicit expressions for all 5 traveling
waves shown by arrows, then using boundary conditions (98) and (99) in each of points x1,2 to get a
system of 4 linear equations, and then solving it.
a
W  x  x1  W  x  x 2 
E Fig. 2.16. A double-barrier system. The

dashed lines show (schematically) the
position of metastable energy levels.
x1 x2 x
However, the transfer matrix approach simplifies the calculations, because we may immediately
use Eq. (114) to write
1  i  i  e ika 0 1  i  i 

T  T Ta T    . (2.121)
 i 1  i  0 
e  ika  i 1  i 
Let me hope that the reader remembers the “row by column” rule of the multiplication of two square
matrices;22 using it for two last matrices, we reduce Eq. (121) to
1  i  i  e ika (1  i ) e ika (i ) 

T    . (2.122)
 i 1  i  e  ika (i ) e  ika (1  i ) 
 
Actually, according to see Eq. (110) we need only one element, T11, of the total matrix:
1 1
T 2
 2
. (2.123)
T11  e  ika  (1  i ) 2 e ika
2
This result is very similar to the solution of Homework Problem 2.3: the transparency is a -periodic
function of the product ka, reaching the maximum (T = 1) at some point of each period– see Fig. 17a.
However, the result is different in that for  >> 1, the resonance peaks of transparency are very narrow,
peaking at ka  kna  n, with n = 1, 2, …
21 Observing that particular matrices (117) and (120) exhibit a substantial degree of symmetry, one can wonder
what of these symmetries are valid for an arbitrary scatterer. Two of the three universal equations relating its
elements may be found from the fact that the probability current should be conserved (j1 = j2) for any combination
of incident wave amplitudes, but the third one requires a bit more refined discussion – see Exercise 6.
N
22 In the analytical form: AB  
jk A
l 1
jl Blk , where N is the matrix dimension (in our current case, 2).
Quantum Mechanics
Physics of this result is immediately clear from the comparison of this result with our analysis of
the simplest quantum well – see Fig. 1.7 and its discussion. At k  kn, the incident wave, which
undertakes multiple sequential reflection from the semi-transparent “walls” of the well, forms a nearly
standing wave which virtually coincides with one of eigenfunctions of the well with infinite walls ( 
). This is the famous effect of resonant tunneling,23 in mathematical description identical to the
resonant transmission of light through an optical Fabry-Perot resonator formed by two parallel semi-
transparent mirrors.
(a) (b)
  0.3 2  2 1
0.8
Im 1
 1
0.6 k
T 1.0 0 Re
0.4
0.2
Fig. 2.17. (a) Resonant tunneling through a
3.0 quantum well with -functional walls and (b)
0 calculating the resonance FWHM at  >> 1.
0 0.5 1 1.5 2
ka / 
Probably, the most surprising feature of this system is the fact that its maximum transparency,
reached at a certain point within each oscillation period, is perfect (Tmax = 1) even at   , i.e. in the
case of very low transparency of each of two component barriers.24 The explanation of this feature is
similar to that of the usual resonance – say, in a mechanical pendulum or an LC tank circuit. Though the
low transparency of the left barrier make the effect of the incident wave on the standing wave inside the
resonator weak, but the same low transparency ensures that the resonator losses are small, i.e. its is
called the quality factor (or just “Q factor”) is high.
Let us find the Q-factor in most interesting limit  >> 1. The Q-factor of n-th resonance peak
may be defined as
kn
Qn  , (2.124)
k
where k is the so-called FWHM, meaning the “full width (of the resonance curve) at half-maximum”,
i.e. the difference k = k+ - k- between such two points on the opposite slopes of the same resonance, in
which T = (Tmax + Tmin)/2 (see arrows in Fig. 17a). The easiest way to calculate k is shown in Fig. 17b.
The denominator in Eq. (123) may be interpreted as the square of difference between two vectors, one of
length 2, and another of length (1 - i)2 = 1 + 2, with angle  = 2ka + const between them. At the
resonance, the vectors are aligned, and the difference is smallest (equals one) and, for  >> 1, is much
smaller than the length of each vector. In order to double the distance squared, the arc, 2, between the
23 Sometimes also called the Ramsauer (or Townsend, or Ramsauer-Townsend) effect. However, it is more
common to use these names for a similar 3D effect, especially at scattering of low-energy electrons on rare gas
atoms – this is how it was first observed in the early 1920s.
24 The statement Tmax = 1 is correct only if both component barriers have an exactly similar transparency.
Quantum Mechanics
vectors should also become equal 1, i.e. 2(2ka + const) = 1. Subtracting the two equations from
each other, we finally get
1
ka  (k   k  )a   1. (2.125)
2
so that, finally,
kn k a
Qn   n  n 2  1 . (2.126)
k ka
(Notice that since, according to Eq. (101),   1/k, the Q factor is actually decreasing as the resonance
number n grows, if well parameters m, W, and a are fixed.)
From classical mechanics we know25 that high Q factor leads to a Q-fold increase of the forced
oscillation amplitude. Similarly, the perfect transparency at resonant tunneling may be explained by the
accumulation of such a large standing wave inside the high-Q resonator, that even its weak penetration
through the “output” (right) barrier yields a wave equal to the incident one.
Let us use this simple system to discuss one more famous uncertainty relation, that between time
and energy. For that, let us consider what will happen if at some initial moment (say, t = 0) we have put
a quantum particle inside the same double-barrier well (Fig. 16) with  >> 1, and leave it there without
any incident wave. For simplicity, let us prepare the initial state so that it coincides with the ground state
of the infinite-wall well:
2 
 ( x ,0 )   1 ( x )  sink1 ( x  x1 ), k1  . (2.127)
a a
At  = , this is an eigenstate of the system, and from our analysis in Sec. 1.4 we know its time
evolution:
E1 k12  2
 ( x, t )   1 ( x) exp i1t, 1    , (2.128)
 2m 2ma 2
telling us that the particle is still in the well with constant probability P(t) = 1. However, if parameter 
is large but finite, de Broglie wave should slowly “leak out” from the well, so that P(t) would slowly
decrease. Let us develop an approximate theory of this effect, assuming that the leakage does not change
the instant wave distribution inside the well, besides the reduction of P.26 Then we can generalize Eqs.
(127), (128) as follows:
2P
 ( x, t )  sink1 ( x  x1 )]exp i1t . (2.129)
a
making the probability of finding the particle in the well equal to P. But this solution may be presented
as a sum of two traveling waves:
 ( x, t )  A expi (k1 x  1t )  B expi (k1 x  1t ), (2.130)
25 See also CM Sec. 4.1.

26 This almost evident assumption finds its formal justification (in the limit E1 >> , where  is the state lifetime
defined below), in the perturbation theory which will be discussed in Chapter 5.
Quantum Mechanics
with equal magnitudes of their amplitudes and probability currents

P  2  P 
A B  , JA  JB  A k1  , (2.131)
2a m m 2a a
slowly decreasing in time. But we already know from Eq. (102) that at  >> 1 the -functional wall
transparency T  1/2, so that the waves (130), incident on the walls from inside, induce outcoming
waves outside of the well with probability currents (Fig. 18)
1  P 
J R  J L  Tj A  . (2.132)
 2 m 2a a
  vgr
jL jR
E1 Fig. 2.18. A simple model

of the metastable state
decay.
0 vgr t x
Now we can use the 1D version probability conservation law (1.?),

dP
 JR  JL  0 , (2.133)
dt
and Eq. (132) to write
dP  P
 2 J R   P , (2.134)
dt ma 
2 2

where the exponential decay constant is called the metastable state lifetime, and in our simple model of
this (very general) phenomenon is equal to
ma 2 2
 . (2.135)

A nice view on this result may be achieved by returning to the resonant tunneling problem and
expressing the resonance width (125) in terms of incident particle’s energy:
 2k 2   2 k1  2 k1 1  2
E     k   . (2.136)
 2m  m m a 2 ma 2 2
Comparing Eqs. (135) and (136), we get a remarkable result
E     . (2.137)
This so-called energy-time uncertainty relation is certainly more general than our simple model; for
example, it is valid for the metastable state lifetime and resonance tunneling width of any metastable
state. This seems very natural, since because of the energy identification with frequency, E = , typical
for quantum mechanics, Eq. (137) may be rewritten as  = 1 and seems to follow directly from the
Fourier transform in time, just as the Heisenberg’s uncertainty relation (1.34) follows from the Fourier
Quantum Mechanics
transform in space. In some cases, these two relations are indeed interchangeable; for example it is easy
to show that Eq. (10) for the Gaussian wave packet width may be rewritten as Et  /2, where E is
the r.m.s. spread of energies of monochromatic components of the packet, while t is the time of the
packet passage through a fixed coordinate x.
Unfortunately, there is a trend, especially among electrical engineers, to exaggerate the
generality of Eq. (137). One can hear, for example, wrong claims27 that due to this relation, energy
dissipated by any system performing an elementary (1-bit) calculation during time interval t has to be
larger than /t.28 Besides insufficient physics education of some authors, there is a deeper reason
for these errors: the coordinate-momentum and energy-time uncertainty relations are actually very much
different. Indeed, in nonrelativistic quantum mechanics, coordinates (say, x), components of the
momentum (say, px), and energy E are actual observables, presented by operators. In contract, time is
treated as a c-number argument, and is not presented by an operator, so that Eq. (137) cannot be derived
as generally as we will (in Chapter 4) derive Eq. (1.34).
We may also use our metastable state picture for one more preliminary discussion of quantum
measurements. Figure 18 shows (schematically) one of the traveling wave packets emitted by the
quantum well after its state (127) was prepared at t = 0. (The similar packet is emitted to the left.) At t
>> , the well is essentially empty (P << 1), and the whole probability distribution rests in two clearly
separated wave packets, with probability 50% each. Let us assume that during a single experiment we
have measured that the particle is actually on the left side of the well. Due to the wide separation of the
packets, this may be evidently done without any actual physical effect on another wave packet.29 But if
we know that the particle has been in the left, there is no chance it has been in the right packet. This
means that the wave function within the right packet should instantly turn into zero- the so-called “wave
packet reduction”, clearly not described by either Schrödinger equation or any other law of physics we
know about.
If we attribute the wave function to each experiment, this may be completely confusing.
However, if (as I already discussed in Chapter 1) we attribute it to a statistical ensemble of similar
experiments, these is no paradox here at all. While the 2-packet picture we have calculated describes the
full initial ensemble (regardless of the particle detection results), the reduced packet picture (with no
wave packet on the right of the well) describes only a specific part of that ensemble, that of experiments
with the particle detected on the left side. As was discussed on completely classical examples in Sec.
1.3, for such sub-ensemble the probability distribution (and hence the wavefunction) may be
dramatically different.
27 See, e.g., K. Likharev, Int. J. Theor. Phys. 21, 311 (1982) for a constructive example that energy dissipation in
may be lower that the claimed “quantum limit” at the so-called reversible computation (introduced in 1973 by C.
Bennett).
28 Another controversial statement is that the energy of a system cannot be measured, during time t, with an
accuracy better than /t – see, e.g., a detailed discussion in: V. Braginsky and F. Khalili, Quantum
Measurement, Cambridge U. Press, 1992.
29 A very convincing argument is waiting for the packets to be separated by such a large distance L = 2vgrt (Fig.
18) that the particle detection may be accomplished in time less than L/c, where c is the speed of light in vacuum,
i.e. the maximum velocity of any physical communication.
Quantum Mechanics
2.5. Coupled systems and band theory

Let us now move on to tunneling through a more complex potential profile shown in Fig. 19: a
sequence of (N – 1) similar quantum wells limited by N similar short tunnel barriers.
a a
E
Fig. 2.19. Resonant tunneling through a
sequence of several similar quantum
wells.
x1 x2  xN x
According to Eq. (114), its transfer matrix is the following product

T  T Ta T ...Ta T , (2.138)

( N 1)  N terms
with the component matrices given by Eqs. (117) and (120), and the barrier height parameter  defined
by the last of Eqs. (101). Remarkably, such multiplication may be carried out analytically,30 giving
1
  sin ka   cos ka  
2
 cos Nqa   
2 2
T  T11 sin Nqa   , (2.139)
  sin qa  
where q is a new parameter, with the wave number dimensionality, defined by the following equation:
cos qa  cos ka   sin ka. (2.140)
For N = 1, Eqs. (139) and (140) immediately yield our old result (94), while for N = 2 they may
be reduced to Eq. (123) and hence illustrated by Fig. 17a. Figure 20 shows its predictions for several
larger numbers N, and several values of parameter . The first surprise comes at N = 3, i.e. two coupled
quantum wells: the resonance peak splits in two, clustered together instead of being evenly spread over
the periodicity interval (ka) = . The splitting is small for larger  (i.e. weak coupling between the
wells) and increases for smaller , but approaches half a period only asymptotically, at   0. The
further increase of N results in the corresponding increase of the number of peaks per period, and at
large well number the peaks merge into a single “energy band” of high transparency, separated from
similar bands in the adjacent periods of function T(ka) by “bandgaps” in which T  0. This is our first
glimpse at the famous energy band/gap pattern which is particular in the basis of all the solid state
theory (and as extension, for all semiconductor electronics). Notice the following important features of
the pattern:
(i) at N  , the band/gap edges become sharp for any , and they tend to fixed positions
(defined by );
30 It will be convenient for me to demonstrate this calculation later on, after we have discussed properties of the
Pauli matrices.
Quantum Mechanics
(ii) the larger interwell coupling (  0), the broader the energy bands and narrower the gaps
between them.
N 3 N  10
0.8   0.3 0.8

) )
0.6 0.6

T T 0.3
0.4 1.0 0.4
1.0
0.2 3.0 0.2
3.0
0 0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
ka /  ka / 
N 5
0.8
)   0.3 Fig. 2.20. Resonance tunneling through N
0.6
equidistant tunnel barriers. Since the
T
0.4 1.0 function T(ka) is -periodic, only one
period is shown.
0.2 3.0
0
0 0.2 0.4 0.6 0.8
ka / 
In order to comprehend the physics of these results, let us analyze an auxiliary system shown in
Fig. 21:31 two similar quantum wells confined by infinitely high potential walls, but coupled via a
transparent, short tunnel barrier (which may be again modeled by a -function):
  , for x   a,

U ( x)  W ( x), for  a  x   a, (2.141)
  , for a  x.

We already know that the standing-wave eigenfunctions n of the Schrödinger equation in regions with
U(x) = 0, in our current case, segments –a < x < 0 and 0 < x < +a, are linear superpositions of sin kx and
cos kx. In order to immediately satisfy the boundary conditions  = 0 at x = a, we can take these
solutions in the form
C  sin k ( x  a ), for  a  x  0,
 n ( x)   (2.142)
 C  sin k ( x  a ), for 0  x   a,
31 This system will be also be our point of entry into the discussion of another topic of key importance, spin and
spin-like systems, in Chapter 6.
Quantum Mechanics
U ( x)
W ( x)
A
EA
S Fig. 2.21. Two lowest eigenfunctions and
ES eigenenergies of a system of two coupled
quantum wells – schematically.
a 0 a x
What remains is to satisfy the boundary conditions at x = 0. Plugging Eq. (142) into Eqs. (98) and (99),
we get the following system of two linear equations:
C  sin ka  C  sin ka,
2mW (2.143)
k (C   C  ) cos ka  C  sin ka.
2
The system has two series of solutions, with two lowest eigenfunctions sketched in Fig. 21:
(i) Antisymmetric solutions (which will be denoted with index A):
C   C  , i.e.  A  C A sin k A x, for both  a  x  0 and 0  x   a, (2.144)
with
sin k A a  0, i.e. k A a  kna  n, n  1,2,... (2.145)
Notice that these values of k, and hence eigenenergies of the antisymmetric states,
 2 k A2  2 n 2
EA   , (2.146)
2m 2ma 2
coincide with those of the simple quantum well of width a – see Fig. 1.7 and its discussion.
(ii) Symmetric solutions (index S):
C   C  , i.e.  S  C S sin k S ( x  a ) , for both  a  x  0 and 0  x   a, (2.147)
with kS satisfying the following characteristic equation:

1
tan k S a   . (2.148)

Figure 22 shows the graphics solution of this equation for three values of parameter , i.e. for various
degrees of quantum well coupling. Each solution kSa is confined within interval
Quantum Mechanics

n  k S a  n  , (2.149)
2
and hence the antisymmetric and symmetric states alternate on the scale of k (and hence eigenenergy),
and the difference kA - kS is below /2a for any value of .
tan ka ,
kA
1 kS


0
 3
 1 Fig. 2.22. Graphics solution of the
characteristic equation for the
eigenvalue of ka in the symmetric
mode, considering parameter 
   0 .3 independent of ka. The dashed line
4 shows approximation (?).
0 1 2
ka / 
Physics of the splitting between eigenenergies corresponding to the symmetric and
antisymmetric states is very simple: it is the change of kinetic energy of the particle due to different
quantum confinement. In each antisymmetric mode, n (0) = n (a) = 0, i.e. the wavefunction is
essentially confined within a segment of length a; as a result, its energy (146) is relatively high. On the
contrary, in the symmetric mode the wavefunction changes slower (cf. two traces in Fig. 21), and hence
its energy is also lower that that of the adjacent antisymmetric mode.
By the way, this problem may serve as a toy model of the strongest (and most important) type of
atom cohesion - the covalent (or “chemical”) bonding in molecules, liquids, and solids. The classical
example of such bonding is that of hydrogen atoms in a H2 molecule. Each of two electrons of this
system32 reduces its kinetic energy very substantially by spreading its wavefunction around both nuclei
protons., rather that being confined near one of them as it has to be in a single atom. As a result, the
bonding is very strong: in chemical units, 429 kJ/mol, i.e. 18.6 eV.33 This is substantially stronger than
the strongest classical (ionic) coupling which is due to electron transfer between atoms, leading to the
Coulomb attraction of the resulting ions. (For example, the atom cohesion of atoms in a NaCl molecule
is just 3.28 eV.)
In the limit   0 (no partition between the wells), kSa  (n - 1/2), i.e. the eigenstates
approach the shape and energy of symmetric states of a quantum well of width 2a. In the opposite limit
 >> 1, kSa  n, and in the vicinity of each such point we can approximate tan kSa with (kSa - n) –
see the dashed line in Fig. 22. As a result, the characteristic equation (148) yields
1
k S a  n  , (2.150)

32 Due to the opposite spins, the Pauli principle allows them to be in the same orbital ground state – see later in
the course.
33 Unit reminder: 1 kJ/mol = 0.0434 eV.
Quantum Mechanics
so that the “splitting” between the wave numbers and eigenenergies of the adjacent symmetric and
antisymmetric states is small:
k A  k S   n 1  2 E n  E n .
2
1 dE
k A  kS   k n , 2 n  E A  E S  (2.151)
a dk ma  a n
Let us analyze properties of the system in this limit in much more detail, because the results will
help us to develop the so-called tight binding approximation in the band theory. Let us focus on a couple
of a symmetric and antisymmetric states, corresponding to the same En. According to Eqs. (144) and
(147), in this limit the system eigenfunctions may be approximately presented as follows:
1 1
 S ( x)   R ( x)   L ( x),  A ( x)   R ( x)   L ( x), (2.152)
2 2
where R,L are the normalized ground states of the insulated wells:
 0, for  a  x  0,  2 / a sin k n x, for  a  x  0,
 R ( x)    L ( x)   (2.153)
 2 / a sin k n x, for 0  x   a,  0, for 0  x   a.
Let us put the particle, at t = 0, into state R(x), and leave the system alone to evolve. Solving Eqs (152)
for R, we may present the initial state as a linear superposition of eigenfunctions:
1
n ( x,0)   R ( x)   S ( x)   A ( x) . (2.154)
2
Now, according to the general solution (1.51) of the time-dependent Schrödinger equation, time
dynamics may be obtained by just multiplying the eigenfunctions by factor (1.45):
1   ES   E A  E  ES
n ( x, t )   S ( x) exp i  t    A ( x) exp i  t  En  A ,
2     2
with (2.155)
1   n    n   En  E  ES
  S ( x) expi  t    A ( x) exp i  t  exp i  t , n  A .
2       2
Now using Eqs. (151) and (152), this expression may be rewritten as
     E 
n ( x, t )   R ( x) cos n t  i L ( x) sin n t  exp i n t . (2.156)
      
This result means, in particular, that the probabilities PR and PL to find the particle, correspondingly, in
the right and left wells change with time as
n n
PR  cos 2 t , PL  sin 2 t, (2.157)
 
mercifully leaving the total probability constant PR + PL = 1. (If our calculation had not passed this
sanity check, that would be a sign of big trouble.)
These are the famous periodic “quantum oscillations” of the particle between two quantum
wells, with frequency  = 2n/, due to their coupling through via tunneling through the tunnel barrier.
Quantum Mechanics
In our current case, the frequency equals En/,34 but the phenomenon is evidently general. It physics of
this phenomenon is straightforward: just as in the single well problem discussed in Sec. 4, the particle
initially placed into a quantum well tries to escape from it via tunneling through is wall. However, now
the particle can only escape into the adjacent well. Trying to escape that well, particle come back, etc.,
just as a classical pendulum, initially deflected from its equilibrium position.
Notice that our everyday intuition, saying that the system should go to an equilibrium situation
(with PL = PR = ½) and stop, is wrong both for the quantum system under analysis and its mechanical
analog, the pendulum, because we treat both systems as conservative (Hamiltonian) ones. We will see
later in the course that as soon as we allow the energy to be drained from the system (via its coupling to
the multi-particle environment), the system eventually indeed tends to static equilibrium, (partly)
justifying the expectations based on everyday experience.
We will prove later that Eq. (156) is valid for (and hence the quantum oscillations take place in)
any system of weakly coupled quantum wells. Let us find En and n for such a system, assuming only
that the potential is symmetric: U(x) = U(-x) – see Fig. 23. The quasi-localized wavefunctions R(x) and
L(x) = R(-x) and their eigenenergies may be found approximately by solving the Schrödinger
equations in one of the wells, neglecting tunneling through the barrier, but finding n requires a little bit
more care.
U ( x)
n
En Fig. 23. Weak coupling between two
similar quantum wells.
a 0 a x
Let us write the stationary Schrödinger equations for the symmetric and antisymmetric solutions
(161),
 2 d 2 S d 2 A
2
E S  U ( x) S  , E A  U ( x) A    , (2.158)
2m dx 2 2m dx 2
multiply the former equation by A, the latter one by S, subtract them from each other, and integrate
the result from 0 to .
 
 2  d 2 S d 2 A 
( E S  E A )  S A dx     2
 A  2
 S  dx. (2.159)
0
2m 0  dx dx 
Now we can apply to this equation the 1D form of the mathematical identity (1.62) to get
34 Notice that the -scaling of the oscillation frequency (  -1) is different that that of the reciprocal lifetime of
the metastable state, given by Eq. (135): -1 -2. We will come back to this fact later in the course (Ch. 5).
Quantum Mechanics
 
 2  d S d A 
( E S  E A )  S A dx    A  S . (2.160)
0
2m  dx dx 0
So far, this is an exact equation. For weakly coupled wells, we can do more. In this case, the left hand
side is very close to -n  (ES – EA)/2, because the integral is dominated by the vicinity of point a, where
the second terms in each of Eqs. (152) are negligible, and the integral is equal to ½, due to the proper
normalization of function R(x). In the right hand side, the substitutions at x =  vanish (due to the
wavefunction decay within the classically forbidden region), and so does the first term at x = 0, because
for the antisymmetric solution A(0) = 0. As a result, we get
2 d A
n   S (0) (0). (2.161)
2m dx
Let us try this formula on the system shown in Fig. 21, with a strong -functional barrier ( >>
1). In this limit, Eqs. (147) and (150) yield
1 1 1
 S (0)  C S sin k S a  C S  , (2.162)
 a
while according to Eq. (144)
d A 1 n
(0)  C A k A  , (2.163)
dx a a
so that Eq. (161) gives
n 2
n  , (2.164)
2ma 2
and thus returns us to the second of Eqs. (151).
In the opposite limit of smooth tunnel barrier, which may be treated within the WKB
approximation, Eq. (161) reduced to35
 1 c 
x
En
n  exp   ( x)dx  , (2.165)
2    xc 
where function (x) is defined by Eq. (70).36

We can now readily interpret the results for resonance tunneling through the double quantum
well system. The splitting between the symmetric and antisymmetric modes clearly manifests itself on
the top left panel of Fig. 20 (for N = 3): each resonance peak corresponds to the incident particle energy
alignment with one of the energy levels. This also gives us a hint how to interpret the resonant tunneling
results for N > 3: if (N - 1) quantum wells are weakly coupled by tunneling through the tunnel barriers
35 See, e.g., the last part of the solution of Problem 4 after Sec. 50 in Landau and Lifshitz, which requires the
simple calculation described in that book just before Eq. (48.3).
36 Comparing this result with Eq. (92), we can notice that again, just as in the case of the -functional barriers,
the transmission coefficient of a barrier scales as a square of the level splitting it provides when coupling discrete
quantum states.
Quantum Mechanics
separating them, there are (N – 1) eigenstates which differ mostly by the phase shifts  between them,
and providing the corresponding small splitting between the energy levels. Now, what are the phase
shifts? It is natural to expect that at the upper level, all  =  (thus providing the highest quantum
confinement), with ka  n at   , while at the lowest level all  = 0, providing the most loose
confinement.37 However, what about  for other levels?
The answer may be readily obtained for an infinite (N  ) periodic structure which is a very
good approximation for crystals, with ~107 atoms in each direction of a few-mm-size sample. It is
almost self-evident that in this case, due to the translational invariance of U(x),
U ( x  a)  U ( x), (2.166)
 should be constant, i.e.
 ( x  a)   ( x) expi  . (2.167)
(A reasonably fair classical image of  is the angle between similar objects - e.g., similar paper clips -
attached at equal distances to a long, uniform rubber band. If the band’s ends are twisted, the twist is
equally distributed between the structure’s periods, representing the constancy of ..)
Equation (167) is the (1D version of) the famous “Bloch theorem”38 which is valid for quantum
particle’s motion in any potential U(x) which satisfies the periodicity condition (166), and states that any
eigenfunction satisfies this equation.. It will be convenient for us to prove the Bloch theorem later on (in
Chapter 4), but I would like immediately to give you its different form. Let us present the real number
 in the form qa (there is no loss of generality here, because parameter q may depend on a as well as
other parameters of the system),39 so that the Bloch theorem takes the form
 ( x  a)   ( x) expiqa. (2.168)
Let us define another function
u ( x)   ( x) exp{iqx} , (2.169)
and study its periodicity:
u ( x  a)   ( x  a) exp{iq ( x  a )}   ( x) exp{iqa}  u ( x) . (2.170)
We see that the new function is a-periodic, and hence we can use Eq. (169) to rewrite the Bloch theorem
as
 ( x)  u ( x) expiqx, with u ( x  a)  u ( x) . (2.171)
37 This expectation is implicitly confirmed by Fig. 20: at  >> 1, the highest resonance peak in each group tends
to ka = n, and the lowest peak also tend to a position independent of N (though dependent on ).
38 After American physicist F. Bloch who applied this concept to wave mechanics in 1928, i.e. very soon after its
formulation. Actually, in mathematics, an equivalent statement had been known before that, and is frequently
called the Floquet theorem, after French mathematician G. Floquet.
39 Parameter q, or rather the product q which has the dimensionality of momentum, is called the quasi-
momentum or (especially in the solid state theory) the crystal momentum of the system. Its most straightforward
interpretation is the average momentum of the particle moving in the periodic potential, i.e. in the conditions
when its genuine momentum p is not conserved.
Quantum Mechanics
We can immediately notice the following key property of Eq. (171): if some pair {u(x}, q} is
used a solution to the Schrödinger equation with some periodic U(x), then the following pair, {u(x), q +
2/a} is also a legitimate solution to the same equation. (Indeed, the replacement qx  (q + 2/a)x = qx
+ 2x/a only multiplies u(x) by a 2-periodic term.) Hence all observables of the system should be 2-
periodic in qa.
Now let us explore the application of the Bloch theorem to find eigenfunctions and eigenenergies
for a particular, and probably the simplest periodic function U(x): an infinite set of similar quantum
wells separated by -functional tunnel barriers (Fig. 24). To start, consider two points separated by
distance a: one of them just left of position xj of one of the barriers, and another just left of the following
barrier (xj+1). Eigenfunctions in each of the points may be presented as linear superpositions of two
simple waves exp{ikx}, and amplitudes of their components should be related by a 22 transfer matrix
T of the potential fragment separating them. According to Eq. (114), this matrix may be found as the
product of matrix (120) of one well and matrix (117) of one barrier:
 A j 1   A   ika 0 1  i  i  A j 
   Ta T  j    e  . (2.172)
B 
 j 1 
B  
 j  0 e  ika  i 1  i  B j 
a a a
En
Fig. 2.24. An infinite set of similar,
equidistant -functional tunnel barriers.
 xj x j 1  x
However, according to the Bloch theorem (167), the component amplitudes should be also related as
 A j 1   A   iqa 0  A j  .
   expiqaI  j    e (2.173)
B  B   iqa  B j 
 j 1   j  0 e  
The condition of self-consistency of these two equations leads to the following characteristic equation:
 e ika
 0 1  i  i   e iqa 0   0 .
  (2.174)
 0
 e  ika  i 1  i   0 e
iqa 
 
Above, we have already calculated the matrix product in this equation – see Eq. (122). Using it, we see
that Eq. (174) is reduced to the same simple Eq. (140) which has already jumped out from the solution
of the different (resonant tunneling) problem.
Let us explore that result. First of all, the right hand part of Eq. (140) is a sinusoidal function of
ka, with amplitude (1 + 2)1/2 – see Fig. 25, while its left hand part is a sinusoidal function of qa with
amplitude 1. As a result, within each period (ka) = 2, the characteristic equation does not have a real
Quantum Mechanics
solution for q inside two intervals of ka (and hence inside two intervals of energy E = 2k2/2m). These
intervals are called energy gaps,40 while the complementary intervals of ka and E, where a real q exists,
are called the allowed energy bands. In contrast, parameter q can take any values, so it is more
convenient to plot the eigenenergy as the function of q (or, even more conveniently, qa) rather than ka.
gap gap …
2
band band …
 1
1
cos qa 0
Fig. 2.25. Graphical representation of the
characteristic equation (140) for a fixed value of
1 parameter . The ranges of ka which yield with cos
qa < 1 correspond to allowed energy bands, while
2
those with cos qa > 1, to gaps between them.
0 1 2 3 4
ka / 
Figure 26 shows these plots for our problem, for the same 3 values of parameter  as were used
in Fig. 20. The band structure of the energy spectrum is apparent. Another evident feature is the 2-
periodicity of the pattern, which we have already predicted from the general Bloch theorem arguments.
(Due to this periodicity, the complete band/gap pattern may be studied on interval -  qa  + , called
the 1st Brillouin zone – the so-called reduced zone picture. For some applications, however, it is more
convenient to use the extended zone picture with -  qa  +. - see, e.g., the next section.)
However, maybe the most surprising fact, clearly visible in Fig. 26, is that there is a infinite
number of energy bands, with different energies En(q) for the same value of q. Mathematically, we could
readily expect it from Eq. (140) – see also Fig. 25. Indeed, for each value of qa there are two solutions
ka to this equation on each period (ka) = 2 - see also panel (a) in Fig. 26. Each of these solutions,
corresponding to a particular energy band, gives a different value of E = 2k2/2m.) However, it is
important for us to understand the physics of these different solutions (and hence energy bands). For
that, let us explore analytically two different limits for the particle energy E. An important advantage of
this approach is that both analyses may be carried out for an arbitrary periodic potential U(x).
(i) The so-called tight-binding approximation is applicable when the energy E is much lower
than the height of the potential barriers separating the potential minima (serving as quantum wells) – see
Fig. 27. As should be clear from our previous discussion that in this limit, the wavefunction is mostly
localized in, and close to at the classically allowed regions at points xj of the potential energy minima
(see dashed lines in Fig. 27). The only role of coupling of these quantum well states (via tunneling
through the separating barriers) is to establish certain phase shifts  between adjacent quasi-localized
wavefunction “lumps” u(x - xj).
To describe this effect quantitatively, let us first return to the problem of only two coupled wells,
and present result (156) in a different but equivalent form:
40 In solid state (especially semiconductor) physics and electronics, term “bandgaps” is more common.
Quantum Mechanics
1st Brillouin zone 1st Brillouin zone

(b) (a)
200 2
 1
 1
150 1.5
E
)100 ka 1
2
50 0.5
0 0
1  0.5 0 0.5 1 1  0.5 0 0.5 1
qa / 2 qa / 2
(c)
200
  0.3
150
E
)100 4 4
E4
2 3
50
0
1  0.5 0 0.5 1
qa / 2
(d)
200
 3
Fig. 2.26. (a) The “real” momentum k of the
150 particle and (b)-(d) its energy E as functions of
E the quasi-momentum q, for the periodic
)100 potential shown in Fig. 23, for the same 3
values of  which were used in Fig. 2.20.
(Energy is plotted in the units of 2/ma2.)
50
Arrows in panel (c) illustrate definitions of the
energy widths of bands (2n) and gaps (2n).
0
1  0.5 0 0.5 1
qa / 2
a a
U ( x) un (x  x j ) u n ( x  x j 1 )
n n
En Fig. 2. 27. Tight binding
approximation (schematically).
0 x j 1 xj x j 1 x
Quantum Mechanics
 E 
n ( x, t )  a R (t ) R ( x)  a L (t ) L ( x)exp i n t , (2.175)
  
where functions aR and aL oscillate sinusoidally in time:
n n
a R (t )  cos t , a L (t )  i sin t. (2.176)
 
It is evident that this evolution may be obtained from the following system of two Schrödinger-like
equations:
da R da
i   n a L , i L   n a R . (2.177)
dt dt
Later in the course (in Ch. 5) we will show that such equations are indeed valid (in the tight-binding
approximation) for any system of two coupled quantum systems. A large advantage of Eqs. (177) is that
they are also valid for the nonstationary case when the well coupling changes in time. What is more
important for us now is that this equation may be readily generalized to the case of many similar
coupled wells. Here, instead of Eq. (175) we should write
 E 
n ( x, t )  exp i n t  a j (t )u n ( x  x j ) , (2.178)
   j
where En are the eigenenergies, and un eigenfunctions of an isolated well. In the tight binding limit, only
the adjacent wells are coupled, so that instead of Eq. (177) we may write an infinite system of similar
equations
da j
i   n a j 1   n a j 1 , (2.179)
dt
with arbitrary integer j, where parameters n still may be found from Eq. (161). Let us solve the
eigenproblem for this system, using the natural assumption
  
a j (t )  expiqx j  i t  const  , (2.180)
  
so that coefficients aj(t), and hence eigenfunction (178), satisfies the Bloch theorem:
a j 1 (t )  a j (t ) exp iqa. (2.181)
Equation (180) assumes that due to the (weak) interwell interaction, system energy E differ from En by a
small amount . To find this amount, it is sufficient to plug Eqs. (180) and (181) into Eq. (179) and
cancel the common exponents. The result is
E  E n    E n   n exp{iqa}  exp{iqa}  E n  2 n cos qa. (2.182)
This result, which corresponds to our numerical plots (see Fig. 26c), is wonderful! Most
importantly, it explains what the energy bands really are – they are just the isolated well levels En,
broadened into bands by the interwell interaction. Second, this gives a clear proof that the energy band
extremes correspond to q = 0 and /a, i.e. to either zero phase shift  between the localized well
functions, or to  = . Finally, the tight binding approximation (whose validity is restricted to n
Quantum Mechanics
<< En) gives a dynamic equation (179) which allows to get many results without specifying the
localized functions u(x).
(ii) Weak coupling limit. Surprisingly, the energy band structure is also compatible with a
completely different picture which can be developed in the opposite limit. Let energy E be so high that
U(x) may be treated as a perturbation. Naively, we would have the parabolic dispersion relation between
particle’s energy and momentum. However, if we are plotting energy as a function of q rather than k, we
need to add 2l/a, with arbitrary integer l, to the argument (Fig. 27). Let us show this using the spatial
Fourier series. For the periodic potential energy U(x) this is straightforward:
 2x 
U ( x)   U l " exp i l", (2.183)
l"  a 
where the summation is over all integers l”, from -  to + . However, for the wavefunction we should
show due respect to the Bloch theorem and expand not  (x) itself, but its periodic component u(x):
 2x 
u ( x)   u l ' exp i l ', (2.184)
l'  a 
so that, according to the Bloch theorem,
 2x   2  
 ( x)  expiqx  u l ' exp i l '   u l ' expi q  l '  x . (2.185)
l'  a  l'  a  
The only nontrivial part of plugging its expression into the Schrödinger equation is the calculation of the
product term
 2x 
U ( x)   u l 'U l " expi q  (l 'l" ) . (2.186)
l ',l "  a 
At fixed l’, we may change summation over l” to that over l  l’ + l”, so that l” = l –l’, and write:
 2x 
U ( x)   expi q  l  u l 'U l l ' . (2.187)
l  a  l '
Plugging expression (184) (with l’ replaced by l), and requiring the coefficient of each spatial exponent
to match, we get an infinite system of linear equations for ul:
2
2  2 
q  l  u l   u l 'U l l '  Eu l  0. (2.188)
2m  a  l'
So far, this system is an exact alternative to the initial Schrödinger equation, and is very efficient for fast
numerical calculations, for virtually any coupling strength, though in systems with tight binding it may
require taking into account a very large number of harmonics un.
In the weak coupling limit, i.e. if all the Fourier coefficients Un are small,41 we can complete all
the calculation analytically. Indeed, in the so-called 0-th approximation we can ignore all Un, so that
Eq. (188) gives:
41 Besides the constant potential U0 which, as we know from Sec. 2, may be included into energy in a trivial way.
Quantum Mechanics
2
2  2l 
E  El  q   , (2.189)
2m  a 
regardless of coefficients ul (which should be obtained from the normalization condition). This result
means that the dispersion relation E(q) has an infinite number of similar quadratic branches numbered
by integer l – see Fig. 28.
E ( 2) 2 2
l 3
l 1 l2
2 1 Fig. 2.28. Band picture in the weak

E (1)
coupling case (n << En). Shading
shows the 1st Brillouin zone.
0 1 qa / 2
On any branch, the wavefunction has just one Fourier coefficient, i.e. presents a monochromatic
traveling wave
 2l  
 l  u l exp{ikx}  u l expi q   x . (2.190)
 a  
This fact allows us to rewrite Eq. (188) the equation in a more transparent form
U
l'
l 'l u l '  ( E  El )u l , (2.191)
which may be formally solved for ul:
U l 'l ul '
ul  l'
. (2.192)
E  El
If the Fourier coefficients Un are nonvanishing but small, this formula shows that wavefunctions
acquire other Fourier coefficients (besides the main one, with the index corresponding to the branch
number), but they are all small, besides the special narrow regions near the points where two branches
(189), with some specific numbers l and l’, cross. (This happens when
 2   2 
q  l    q  l' , (2.193)
 a   a 
i.e. at qa  (l + l’), corresponding to
2  2 2 2
El  El '  2
 (l  l ' )  2l 2
 2
n  E (n) , (2.194)
2ma 2ma
Quantum Mechanics
with n  l – l’. Equation (194) shows that index n is just the number of the branch crossing on the energy
scale – see Fig. 28.) In such a region, E has to be close to both El and El’, so that the denominator in
just one of the infinite number of terms in Eq. (192) is very small, making the term substantial despite
the smallness of Un.. Hence we can take into account only one term in each of the sum (written for l and
l’):
U  n u l '  ( E  El )u l ,
(2.195)
U n u l  ( E  El ' )u l ' .
Taking into account that for any real function U(x) the Fourier coefficients are related as U-n = Un*, Eq.
(195) yields the following simple characteristic equation
E  El  U n*
 0, (2.196)
Un E  El '
with solution
1/ 2
 El  El '  2 
E   E ave     2n  , (2.197)
 2  
whose parameter Eave  (El + El’)/2 = E(n) does not depend on the distance from the central point qa = n.
This gives us the famous anticrossing diagram shown (Fig. 29) with the energy gap minimum equal to
2n, where
2n  U nU n*  U n .
2
(2.198)
(These regions are also clearly visible in Fig. 28 which shows the results given by Eq. (140) for  = 0.1)
E  En
E
2 n
0 El  El '
E
Fig. 2.29. The anticrossing diagram
We will run into the anticrossing diagram again and again in the course, notably at the discussion
of spin. Such diagram characterizes any quantum systems with two weakly-interacting eigenstates with
close energies.42 It is also repeatedly met in classical mechanics, for example at the calculation of own
frequencies of coupled oscillators – see, e.g., CM Sec. 4.9. Actually, we could obtain this diagram
42 In recent physics literature, they are called either two-level systems, or spin-like systems, or (especially in the
context of information coding and processing) qubits.
Quantum Mechanics
earlier in this section, for the system of two weakly coupled quantum wells (Fig. 23), if we assumed the
wells to be slightly dissimilar.
In our current case the diagram describes the weak interaction of two sinusoidal de Broglie
waves (189), with oppositely directed wave vectors, l and –l’ , via the (l -l’)-th Fourier harmonic of
potential U(x). This effect exists also for the classical wave theory, and is known as the Bragg reflection,
presenting, for example, the weak-interaction limit of the wave reflection by a crystal lattice – see Fig.
1.5.
Returning for the last time to our initial result – the band structure shown in Fig.. 26, we may
wonder how general is this pattern, obtained for the -functional potential (Fig. 24). A partial answer
may be obtained from the band structure for two other renowned periodic functions U(x): the
rectangular Kronig-Penney potential (Fig. 30a) and sinusoidal (“Mathieu”) potential shown in Fig. 30b.
d (a)
U ( x) A
a
U0
0 x
a Fig. 2.30. Two more periodic potential
U ( x) (b)
A profiles whose band structures have been
studied in much detail: (a) the Kronig-
U0 Penney potential and (b) the sinusoidal
(“Mathieu”) potential.
0 x
I will leave a detailed study of the Kronig-Penney potential for students (see Homework 6). The
Schrödinger equation for the sinusoidal potential,
 2 d 2  2x 
 2
 U 0  A cos   E (2.199)
2m dx  a 
may be reduced to the canonical form of the Mathieu equation43
d 2
 (a  2q cos 2 )  0. (2.200)
d 2
by the introduction of new notation
x E U0 A  2 2
 , a , 2q  , E1  .. (2.201)
a E1 E1 2ma 2
43 Of course, here a and q have nothing to do with our prior use of these letters (for the structure period and
quasi-momentum). Let me hope that this occasional use, which follows the established tradition, will not cause
confusion.
Quantum Mechanics
Fig. 2.31. Characteristic values of the Mathieu

equation. In the application to the band theory,
2q dotted regions correspond to the energy gaps.
The dashed line corresponds to condition 2q =
a, i.e. E = U0 + A  Umax, separating the
regions of the under-barrier and over-barrier
motion.
4a
Figure 31 shows the “characteristic values” of the Mathieu equation, i.e. the relations between
parameters a and q, corresponding to the allowed energy bands and gaps. (In such “phase plane” plots,
the detailed information about the energy dependence on the quasi-momentum is lost, but we already
know from Fig. 26 that the dependence is not too eventful.) The comparison of Figs. 26 and 31 shows
that the main difference provided by relatively smooth potential U(x) is the virtual disappearance of
energy gaps at particle energies exceeding the maximum value of the potential energy. (For the Mathieu
potential this condition is E > U0 + A, i.e. a > 2q – see the dashed line in Fig. 31.) This is natural,
because this is essentially the condition of crossover between particle tunneling through the “humps” of
the potential profile (well described by the tight binding approximation with its narrow allowed bands)
to its virtually free motion (which is much closer to the weak coupling limit with its narrow interband
gaps). On the contrary, the energy gap width in Fig. 26 only grow with energy – an evident artifact of
the -functional model for function U(x), with Umax = .
It is evident that the energy band theory should be valid not for an infinite sample, but also for
sample of a very big length L = Na >> a. (This statement finds an implicit confirmation in the
comparison of plots shown in Fig. 20 for different N.) However, for such sample the eigenstates are
discrete; let us examine how they are distributed along axes q and E.
Presenting eigenstates as standing waves in a quantum well, as was done in Sec. 1.4, is a bit
inconvenient, because we would like to speak about the distribution of traveling-wave states. A useful
trick here is to argue that in a very long sample the situation should not depend on the exact boundary
conditions at its borders,44 and use the “Born-Karman” boundary conditions (after M. Born and T. von
Kármán)
 ( L)   (0) . (2.202)
(This condition may be physically implemented by bending 1D segment [0, L] into a large loop along
other coordinates, and connecting the ends. Of course, this trick requires intuitive (and correct!) belief
that such bending does not affect 1D physics too much, provided that L >> a.)
In the simplest case of free particles, boundary conditions (202) allow traveling-wave solutions
  C exp{ikx} , (2.203)
provided that k takes only the following values:
44 The same argument allows us to limit our discussion to the case when L is an exact multiple of a.
Quantum Mechanics
k l L  2 l , (2.204)
with any integer l. (The physics of this result is of course very simple: the sample of length should
contain an integer number of de Broglie wavelengths  = 2/k). This means that the states are
distributed along axis k uniformly, with equal distances45
2
k  (2.205)
L
between the next neighbors. A popular and convenient way to express this simple result is to say that for
a large sample a sum over the states may be replaced by an integral:
L
 f (k )  2  f (k )dk ,
k
(2.206)
k
where f(k) is any function of k. In particular, if we take f(k) = 1, this relation gives a convenient way to
count the number n of states within a certain interval of k:
L
2 k
n dk . (2.207)
From this result, we can readily calculate the density of free-particle states in energy:
dn dn / dk L / 2 1 1
g1    2   1/ 2 . (2.208)
dE dE / dk  k / m k E
Now, let us return to the case when the periodic potential U(x) is not negligible. Surprisingly, the
Bloch theorem makes the analysis elementary. Indeed, plugging the Bloch wavefunction (171) into
boundary conditions (202), we see that distribution (204) remain valid, for any U(x), for eigenvalues of
the quasi-momentum q. As a result, the state-counting rule is salvaged in the form
L
 f (q)  2  f (q)dq .
q
(2.209)
q
However, the distribution of these states along axis E may be now dramatically different. For example
the 1D density of states
dn dN / dq L / 2
g1    (2.210)
dE dE / dq dE (q ) / dq
diverges at both edges of each allowed energy band (where dE(q)/dq  0 – see, e.g., Figs. 26, 28, 29).
These (integrable) divergences are very important for applications, e.g., for solid state lasers (“laser
diodes”), because they provide an abundance of state pairs with the mutual energy differences very close
to the value of the energy gap 2n, and hence to each other, enabling mutual synchronization of quantum
transitions – the “stimulated emission” of radiation.
45 An attentive reader could have noticed that this distance is twice larger than that (k = /L) following from our
analysis of a 1D quantum well in Sec. 1.4 – see Eq. (1.60). The origin of this difference is that in that analysis,
based on different boundary conditions, we counted standing-wave solutions with k > 0, each of which may be
presented as a sum of two traveling waves with wavenumbers +k and –k. As a result, the total number of states,
within a certain interval of k, is the same in both approaches.
Quantum Mechanics
2.5. Bloch oscillations, Stark ladder, and Landau-Zener tunneling

(FOR NOW, JUST A DRAFT, WITH VERY FEW EXPLANATIONS - KKL)
Important problem: a periodic potential U(x) plus external force F (independent of x, but
possibly depending on time). Example: a crystal in applied electric field. In the WKB approximation
(valid if Fa << n, En for each relevant energy band), the quasi-momentum obeys the 2nd Newton’s
law:
dq
  F (t ) . (2.211)
dt
Bloch oscillations – two pictures. In the time-domain picture (Fig. 32a), motion along q axing
within one energy band. From here, the time period
q 2 / a 2
t B    , (2.212)
q F / Fa
i.e. the Bloch oscillation frequency
2 Fa
B   . (2.213)
t B 
From the energy picture (Fig. 32b), EF = E(q) – Fx, it is easy to derive the spatial amplitude of
the Bloch oscillations (or, in the energy-domain picture, the spatial spread of the eigenfunction):
x B  E n / F . (2.214)
E (q ) (a) EF (b)
21 2 1 / F
21
E1
E1
 0  qa a 0 x
E1 / F
Fig. 2.32. The Bloch oscillations (red lines) and Zener tunneling (blue lines) from two points of
view: (a) time-domain picture, and (b) energy-domain picture.
Experimental situation in solid crystals (metals or semiconductors): bad for realistic fields (too
much scattering, heating). Semiconductor superlattices: virtually perfect structures, much larger a, much
shorter tB – may be shorter than the mean scattering time – realistic!. But even here, generated rf power
is too low (independent contributions by the different electrons). Way to observe: the “Stark energy
ladder” (Fig. 32b), with adjacent eigenfunction shift xS = a and hence the energy level separation
Quantum Mechanics
dE
E S  x S  Fa . (2.215)
dx
Applying additional rf field (typically, shining microwaves on the sample) induces transitions between
the levels – may be observed by resonant absorption. Notice that ES = B, i.e. we are speaking, into
two languages, about the same phenomenon – very typical for physics.
Next question, what happens at the particle reflection from the bandgaps? Let us forget about F
for a second. For weak coupling (narrow-bandgap) limit, Eq. (197) may be rewritten as an explicit E(q)
dispersion law:
E  E   E ave   q   2n 2


1/ 2
, (2.216)
where q  (q - n/a) and

d ( El  El ' )  2 n 4
 2  En , (2.217)
dq El  El '  E n
m a n
so that
q  
1

E  2
 2n 
1/ 2
. (2.218)
Generalizing for the gap, q  i , we get

1

 2
n  E  
2 1/ 2
. (2.219)
1/ shows how deep do de Broglie waves, with an energy inside a gap, penetrate into the system (say,
from an edge).
Now returning to weak field F: now , E = -Fx +const, so that
   ( x) 

1
 2
n  ( F~
x )2 
1/ 2
, (2.220)
and ~
x  0 corresponds to mid-gap, and from the WKB formula (92), we get
1/ 2
4 n c  ~
x
x2  ~ 2n
 ln T  2   ( x)dx  0  xc2 
 1   d x  . (2.221)
2
 ( x ) 0
  F
Plugging in expression (217) for , we can reduce this result to a manifestly dimensionless form
  2 n2n 
T  exp  (2.222)
 4 E n Fa 
which shows that the transparency grows very fast with the applied field. Notice that since the
quasiclassical theory is only valid at Fa << n, En, this expression is only valid if T << 1.
Quantum Mechanics
This is the famous Zener (or “interband”) tunneling. Practical applications: tunnel (“Zener”)
diodes, capable of the negative differential resistance effect (dV/dI < 0), very useful for the amplification
or generation of analog signals, and digital signal swing restoration.
Returning to Eq. (211) for dq/dt, we can also rewrite combination F/ as
F d ( El  El ' ) dq d ( El  El ' )
  V , (2.223)
 dq El  El '  E n
dt dt El  El '  E n
(where V has the clear meaning of the “energy speed” of the particle in the absence of the gap), and
present Eq. (222) in a different form:
 2 
T  exp n  . (2.224)
 V 
Interpretation: since n/V gives the time scale t of energy’s crossing the gap region, our result means
n
 ln T  t . (2.225)

Ratio /t gives scale EV of energy available for the excitation over the gap, so that our result may be
reduced to the usual “activation formula” T = exp{-n/EV}.
The exact calculation of T in the time-domain picture is more complex (was done independently
by L. Landau and C. Zener in 1932),46 and yields
  n 
T  exp . (2.226)
 4 V 
Its advantage of the last result is that it is valid for arbitrary T. In Chapter 5, we will derive this result,
albeit only for T << 1, from the perturbation theory.
2.6. Harmonic oscillator: Brute force approach

To complete our review of 1D systems, we have to consider the famous harmonic oscillation, i.e.
a particle moving in the quadratic-parabolic potential (Fig. 10)
m 02 2
U ( x)  x . (2.227)
2
This is just a smooth quantum well providing “soft” confinement, whose discrete spectrum we have
already found in the WKB approximation – see Eq. (89):
 1
E n   0  n   , (2.228)
 2
46 See also C. Wittig, J. Phys. Chem. B 109, 8428 (2005).
Quantum Mechanics
Now we have to solve the same problem exactly – not because there is anything conceptually
interesting in it, but because of its enormous importance for applications. Let us write the stationary
Schrödinger equation for potential (227):
 2 d 2 m 02 2
  x   E . (2.229)
2m dx 2 2
From the solution of Homework Problem 1.3, we already know one of the eigenfunctions of this
equation,
 m 0 x 2 
 0  C 0 exp , (2.230)
 2 
and the corresponding eigenenergy
 0
E0  . (2.231)
2
From the WKB result (6.2), we may be pretty sure this is the ground state, corresponding to n = 0,
because, as we have seen on several examples, for soft potentials the approximation’s error is typically
below 10% even for the lowest state (hardest for this method).
To find other eigenstates, let us make Eq. (6.3) dimensionless. Equation (6.4) clearly shows how
to do it: the characteristic scale of wavefunction spread:
1/ 2
  
x0    . (2.232)
 m 0 
Its physics meaning is clear from the following identity:
m 02 2  0
U ( x0 )  x0   E0 . (2.233)
2 2
Due to the importance of this scale, let us give its crude estimates for several important systems:
(i) Electrons in fluids, and solids: m  10-30 kg, 0 ~ 1015 s-1, giving x0 ~ 0.3 nm, comparable
with inter-atomic distances a. As a result, classical mechanics is not valid at all.
(ii) Atoms in solids: m  10-26 kg, 0 ~ 1013 s-1, giving x0 ~ 0.1 nm ~ a/3. Because of that,
classical mechanics (e.g., molecular dynamics) is approximately OK, but misses some effects – e.g.,
quantum tunneling of hydrogen atoms through energy barriers of the potential profile created by its
neighbors.
(iii) LIGO47 probe masses: m ~102 kg, 0 ~ 104 s-1, giving x0 ~ 10-22m. After almost two decades
of work (and hundreds of millions of NSF dollars:-), this experiment is still struggling with seismic
noise at the level of a few 10-20 m, and plans for future versions, called “enhanced LIGO” and “advanced
LIGO” promise to decrease the noise by only an order of magnitude. Thus the prospects of observing
quantum-mechanical effects, much heralded at the initial stage of work on these instruments, do not
seem very realistic.
47 Laser Interferometer Gravitational-Wave Observatories, see online at http://www.ligo.caltech.edu/.
Quantum Mechanics
Returning to the Schrödinger equation, introducing dimensionless variable   x/x0, we get

d 2
   2   , (2.234)
d 2
where   2E/0 = E/E0. In this notation, the ground state wavefunction is proportional to exp{-2/2},
so that let us look for the solutions to Eq. (229) in the form
 2
  C exp  H ( ) , (2.235)
 2
where H() is a new function. Substituting Eq. (6.9) into Eq. (6.8), and canceling the exponent, we get:
d 2H dH
 2  (  1) H  0 . (2.236)
d 2
d
It is evident that H = const and  = 1 is one of the solutions, describing the ground state (6.4)-(6.5), but
what are the other eigenstates and eigenvalues? This equation has been studied in detail in the mid-
1800s by French mathematician C. Hermit who has shown that all eigenvalues are given by equation
 n  1  2n, with n = 0, 1,2,…, (2.237)
so that our WKB result (6.2) is indeed exact. The eigenfunctions corresponding to eigenvalue n is a
polynomial (called the Hermite polynomial) of degree n, which may be most conveniently calculated
using the following formula:
H n   1 exp  n exp  2 .
n dn 2
(2.238)
d
It is easy to use this formula to calculate several lowest-degree polynomials:
H 0  1,
H 1  2 ,
H 2  4 2  2, (2.239)
H 3  8 3  12 ,
H 4  16 4 - 48 2  12.
The most important properties of the polynomials (Fig. 33a) are as follows:
(i) their “parity” (symmetry-antisymmetry) alternates with the number n,
(ii) Hn() crosses the -axis n times (has n zeros), and
(iii) the polynomials are mutually orthonormal in the following sense:

H n ( )H n ' ( ) exp  2 d   1 / 2 2 n n! n ,n ' . (2.240)


Quantum Mechanics
10
n2 (a)
n 1
H n ( )
n0
0
n3
 10
3 0 3

3 (b)
E3
2 Fig. 2.33. (a) A few lowest Hermite
polynomials and (b) the corresponding
E2
eigenenergies (dashed lines) and
U ( x) 1 eigenfunctions (solid lines) of the
harmonic oscillator. The dashed-line
E1 parabola shows the potential profile
0 U(x), drawn on the same scale as
energies En, so that the dashed line
E0 crossings correspond to the classical
turning points.
0 x / x0
Translating this result to coordinate x, we can write the following orthonormal eigenfunctions of the
harmonic oscillator (Fig.33b):
1/ 2 1/ 4
 1   m 0   x2   x 
 n ( x)   n    exp 2  H n   . (2.241)
 2 n!      2 x0   x0 
(These stationary states of the harmonic oscillator are sometimes called its Fock states, to distinguish
them from other fundamental states which will be discussed later in the course.)
A typical example of states in a soft-wall quantum well. Instructive to compare with the
rectangular quantum well, with its ultimately-hard walls.
Similar features:
(i) Oscillations in the classically-allowed regions with En > U(x).
(ii) Alternation of symmetric and antisymmetric wave functions with the growth of n.
Different features of the soft confinement:
Quantum Mechanics
(i) A growing spatial spread of the wavefunction, which follows the gradual increase of
the classically allowed region, with n.
(ii) The corresponding slower growth of En, because of the gradual reduction of quantum
confinement, which moderates the growth of kinetic energy. By the way, later we will show that our
particular system, the harmonic oscillator, always keeps both energies equal:
p2 En
 U ( x) n
 . (2.242)
2m n
2
Unfortunately, most formulas discussed in this section, with the notable exception of Eq. (228),
give us both more and less information than we need for applications. (For example, the proof of Eq.
(238) is rather longish. More importantly, it is also hard to use this brute force approach for calculation
of the so-called matrix elements of the system – virtually the only numbers important for applications.)
It is also almost evident that there should be some straightforward math leading to the simple formula
(228) for En. We will develop a much more efficient (“operator”) approach to this problem in Chapter 4.

Chapter 2. 1D Wave Mechanics: X P I X X X X

Uploaded by

Copyright:

Available Formats

You might also like

Chapter 2. 1D Wave Mechanics: X P I X X X X

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 2. 1D Wave Mechanics: X P I X X X X

Uploaded by

Copyright:

Available Formats

Quantum Mechanics

Chapter 2. 1D Wave Mechanics

2.1. Free particle: Wave packets

normalized to P = 1, for any parameters x and p0.2

 exp{ 2 }d   / 2 , may be readily used to prove that the r.m.s.

2.2. Wave mechanics in piecewise-constant potential profiles

classically accessible classically forbidden

U (x) Fig. 2.4. Classical 1D motion in a potential

0.2 Fig. 2.6. (a) Transmission coefficient of a potential

U 0 Fig. 2.7. Rectangular tunnel barrier.

2.3. The WKB approximation

where 0 still follows Eq. (59), while 1 describes a small correction:

Plugging Eq. (62) into Eq. (57), we get

15 The following integral formulas,

we see that the additional phase shift depends on particle’s energy:

Fig. 2.10. Quasiclassical treatment of eigenstates in a

so that Eq. (84) is satisfied if

which conserves the probability current.

On the other hand, the wavefunction itself is still continuous:

Fig. 2.13. A single 1D scatterer.

E Fig. 2.16. A double-barrier system. The

1  i  i  e ika 0 1  i  i 

1  i  i  e ika (1  i ) e ika (i ) 

25 See also CM Sec. 4.1.

with equal magnitudes of their amplitudes and probability currents

E1 Fig. 2.18. A simple model

Now we can use the 1D version probability conservation law (1.?),

2.5. Coupled systems and band theory

According to Eq. (114), its transfer matrix is the following product

0.8   0.3 0.8

with kS satisfying the following characteristic equation:

where function (x) is defined by Eq. (70).36

1st Brillouin zone 1st Brillouin zone

2 1 Fig. 2.28. Band picture in the weak

which may be formally solved for ul:

Fig. 2.31. Characteristic values of the Mathieu

2.5. Bloch oscillations, Stark ladder, and Landau-Zener tunneling

E  E   E ave   q   2n 2

where q  (q - n/a) and

Generalizing for the gap, q  i , we get

2.6. Harmonic oscillator: Brute force approach

46 See also C. Wittig, J. Phys. Chem. B 109, 8428 (2005).

47 Laser Interferometer Gravitational-Wave Observatories, see online at http://www.ligo.caltech.edu/.

Returning to the Schrödinger equation, introducing dimensionless variable   x/x0, we get

H n ( )H n ' ( ) exp  2 d   1 / 2 2 n n! n ,n ' . (2.240)

You might also like