Tunneling Through a Potential Barrier

Douglas H. Laurence

Department of Physical Sciences, Broward College, Davie, FL 33314

1 Introduction
Consider a classical object moving along the ground with some kinetic energy K, when it approaches
a hill. The height of the hill is such that the gravitational potential energy of the object at the top
of the hill would be V . If K > V , the object will climb over the hill and keep moving, but if K < V ,
the object won’t have enough kinetic energy to reach the top of the hill, and will turn around and
move in the opposite direction once it reached a height such that K → 0. This situation is the
classical analogy to a free quantum particle, with some total energy E, approaching a potential
barrier of some height V . The expected result, from a classical analysis, is that the object would
pass through the barrier if E > V , but be turned around by the barrier if E < V . However, this
isn’t the case: there’s a non-zero probability that the particle will still pass through the barrier if
E < V . This is known as quantum tunneling, or simply tunneling.
Tunneling has an actual, observable impact on life. It most commonly is seen in α-decay of
radioactive nuclei. An alpha particle, bound in a nucleus, often never has enough energy to overcome
the strong force binding it. However, it has a change to simply tunnel through the potential barrier
and pop out on the other side a free α particle. This is exactly how this decay process works. This
explanation was proposed by George Gamow in 19281 , and we will cover it at the end of this note.
Tunneling also puts limits on how efficient microprocessors can become. As it currently stands,
the transistors used in computer processors are the so-called “field-effect” transistors (FETs), which
use a controllable electric field (in an “on-off” configuration) to either prevent electrons from moving
through the transistor, or to allow them to move through. The problem is that they require tens
of thousands of electrons to produce a signal currently. The ideal scenario would be to produce a
single-electron transistor, which would only require a single electron to move across the transistor
to produce a signal. However, counting on a single electron to sit inside a transistor, holding it
back with an electric potential barrier, isn’t going to work 100% of the time, since the electron
can tunnel through this barrier. It’s very different in FETs, which require tens of thousands of
electrons to produce a signal, so a single electron tunneling won’t produce a signal, and the odds
of ten thousand tunneling at the same time is essentially zero.

2 Direct Solution of Schrödinger’s Equation

Consider a free particle with energy E > 0 encountering a potential barrier of height V0 > E, which
extends from x = 0 to x = a. This is not a classically passable barrier, since the particle doesn’t
have enough energy to overcome the potential. But we will show that there does exist a non-zero
1
G. Gamow, “Zur Quantentheorie des Atomkernes (On the quantum theory of the atomic nucleus),” Z. Phys, 51,
204 (1928).

1
probability of the particle doing exactly that: passing through the barrier. To begin, let’s divide
the x-axis into three regions:
Region I: −∞<x<0
Region II: 0 < x < a
Region III: a<x<∞
In Region I and III, Schrödinger’s equation is going to be:
d2 ψ 2mE
2
= − 2 ψ = −k 2 ψ (k ∈ R) (1)
dx ~
which is going to have the solutions:

(Region I) ψI (x) = Ae−ikx + Beikx

(2)
(Region III) ψIII (x) = F e−ikx
Imagine that the problem was as depicted in Figure 1, where I am using the convention that
a wave e−ikx moves in the +x-direction. In the figure, it’s pointed out that a wave moving in the
−x-direction in region III is unphysical: consider the right-moving wave in Region I as the wave
indident on the boundary, and the left-moving wave in Region I as the reflected wave, then a right-
moving wave in region III is clearly a transmitted wave, but there is no physical interpretation for
a left-moving wave in region III.

Figure 1: The physical solutions to Schrödinger’s equation for V0 > E.

The transmission coefficient, the probability of passing through the barrier, is just going to be
the relative probability of measuring a right-moving wave in region III vs. a right-moving wave in
region I, i.e.:
|F |2
T = (3)
|A|2
while the reflection coefficient, the probability of bouncing off of the barrier, is going to be the
relative probability of a left-moving vs. right-moving wave in region I, i.e.:
|B|2
R= (4)
|A|2
Note that in neither of the above scenarios is |A| = 1; this is because you can’t normalize free
particle solutions. A relation between transmission and reflection coefficients naturally exists due
to their probabilistic interpretations:
R+T =1 (5)

2
though there exists the same relation in classical optics without the probabilistic interpretation.
Now, in Region II, Schrödinger’s equation is going to be:
d2 ψ 2m(E − V0 )
=− ψ = +κ2 ψ (κ ∈ R) (6)
dx2 ~2
where the inclusion of the negative sign in the definition of κ2 is necessary to make κ real since
E − V0 < 0. The wavefunction in Region II is therefore:
ψII (x) = Ceκx + De−κx (7)
The boundary conditions are going to be that the wavefunction and its derivative are continuous
at x = 0 and x = a. This gives us four equations, though there are 5 unknowns: A, B, C, D,
and F . However, recall that the reflection and transmission coefficients are defined as ratios with
respect to |A|2 , so our four boundary conditions can give the four coefficients B, C, D, and E as
ratios with respect to A, allowing us to find R and T . These four equations are:
ψI (0) = ψII (0)
⇒ A+B =C +D (BCI)

ψI0 (0) = ψII

0 (0)

⇒ −ik (A − B) = κ (C − D) (BCII)

ψIII (a) = ψII (a)

⇒ F e−ika = Ceκa + De−κa (BCIII)
0 (a) = ψ 0 (a)
ψIII II
⇒ −ikF e−ika = κ (Ceκa − De−κa ) (BCIV)

We can use (BCIII) and (BCIV) to solve for C and D as functions of F . First, we want to
divide (BCIV) by κ, then multiply (BCIII) and (BCIV) by eκa , yielding:

F e−ika eκa = Ce2κa + D

ik −ika κa
− Fe e = Ce2κa − D
κ
Adding these equations, and solving for C, gives:
 
1 ik
C= 1− F e−ika eκa (8)
2 κ
We need to use (BCIII) and (BCIV) again to solve for D, doing basically the opposite of before.
We’ll divide (BCIV) by −κ this time, and then multiply (BCIII) and (BCIV) by e−κa , resulting in:

F e−ika e−κa = C + De−2κa

ik −ika −κa
Fe e = −C + De−2κa
κ
And adding these equations, and solving for D, yields:
 
1 ik
D= 1+ F e−ika e−κa (9)
2 κ

3
 Now we want to take these results for C and D and plug them into (BCI) and (BCII). First,
let’s work with (BCI):
   
1 ik −ika κa ik
A+B = 1− Fe e + 1+ F e−ika e−κa
2 κ κ
1 1 ik −κa
= (eκa + e−κa )F e−ika + (e − eκa )F e−ika
2 2κ
 
ik
= cosh(κa) − sinh(κa) F e−ika (EI)
κ

For (BCII), first we’ll divide both sides by κ, and then plug in the results for C and D. We’ll get
a very similar result as above:
   
ik 1 ik −ika κa 1 ik
− (A − B) = 1− Fe e − 1+ F e−ika e−κa
κ 2 κ 2 κ
1 1 ik κ
= (eκa − e−κa )F e−ika − (e + e−κa )F e−ika
2 2κ
 
ik
= sinh(κa) − cosh(κa) F e−ika
κ

Now moving the factor of −ik/κ from the left-side to the right-side, we have:
 
iκ ik
A−B = sinh(κa) − cosh(κa) F e−ika (EII)
k κ

Adding these equations (EI) + (EII) yields a solution for A:

   
κ k
2A = 2 cosh(κa) + i − sinh(κa) F e−ika
k κ

So, we can solve for F as a function of A, exactly as we wanted to when we started:

2e−ika
F = A
2 cosh(κa) + i κk − κk sinh(κa)

Now we need to take the modulus-square of both sides, and solve for T = |F |2 /|A|2 . Notice
something, though: the modulus square is going to involve a e−ika ∗ eika = 1 in the numerator,
and a (A + iB)(A − iB) = A2 + B 2 in the denominator. Because the numerator will simply be
a number, but the denominator will be a more complicated function, I’ll express the transmission
coefficient as an inverse T −1 instead, for convenience. Thus, we get:

1 κ k 2
 
−1 2
T = cosh (κa) + − sinh2 (κa)
4 k κ

There’s one last bit of simplification we need to perform. Recalling that cosh2 (x) = 1 + sinh2 (x)
for hyperbolic trig functions, we see T −1 becomes:
"  #
κ k 2 1 4k 2 κ2 + κ4 + k 4 − 2k 2 κ2
  
−1 1
T =1+ 4+ − sinh2 (κa) = 1 + sinh2 (κa)
4 k κ 4 k 2 κ2

4
Noting that κ4 + k 2 + 2k 2 κ2 = (κ2 + k 2 )2 , we find one of our final solutions for T −1 :
 2
1 κ k
T −1 = 1 + + sinh2 (κa) (10)
4 k κ

This isn’t the only way to express the transmission coefficient, though. Often times, it’s preferred
that it be expressed in terms of the energies E and V0 that set the details of the problem. Recalling
our definitions of k and κ, from equations (1) and (6), we see that:

κ k κ2 + k 2 (V0 − E) + (E) V0
+ = = p =p
k κ kκ E(V0 − E) E(V0 − E)

So, a popular alternate to T −1 given above is:

V02
 
T −1 = 1 + sinh2 (κa) (11)
4E(V0 − E)

The above solution, plotted against E/V0 , is given in Figure 2 below.

Figure 2: T for tunneling. Image credit: WikimediaCommons, author Bamse.

3 Approximate Solutions for Tunneling

First, let’s look at the “weak tunneling” limit, in which κa  1. To see how T −1 should act, let’s
re-write sinh(κa) in terms of exponentials:
1
sinh(κa) = (eκa − e−κa )
2
What’s going to happen in the κa  1 limit? Well, eκa is going to blow up, but e−κa is going
to drop to zero. We don’t want our result to blow up, but keep in mind we’re not analyzing the
transmission coefficient, but the inverse of it. This means in order to keep T from blowing up, we

5
need to drop the e−κa and only keep the e−κa . Even though T −1 will blow up, T will remain finite.
So, we have:
1 k 2 2κa 1 κ k 2 2κa
   
−1 κ
T ≈1+ + e ≈ + e
k 16 κ 16 k κ
where the second approximation is because, as e2κa blows up, it gets much larger than 1, so we can
ignore it. So, in the limit κa  1, the transmission coefficient is:
16k 2 κ2 −2κa
T = e (12)
(κ2 + k 2 )2
Or, expressed in terms of E and V0 , the transmission coefficient is:
16E(V0 − E) −2κa
T = e (13)
V02
Tunneling isn’t typically analyzed the way I’ve done it so far. Typically, tunneling is looked
at using what’s known as the WKB approximation2 . The WKB approximation is used when a
potential barrier encountered is constant or nearly constant. This means that we can, of course,
use the WKB approximation to good effect for tunneling, since the potential barrier is constant
during the process. The approximation that’s actually made in WKB is to assume that if V (x)
varies slowly, relative to some wavelength λ (i.e. the de Broglie wavelength of the particle), then
we assume a solution:
ψ(x) = A(x)eiφ(x) (14)
with φ(x) either real or imaginary, depending on how the energy compares to the potential barrier
at x. Since we’re treating V (x) as varying slowly, we also want to treat A(x) as varying slowly,
such that we ignore all derivatives A(n) (x) for n > 1, i.e. we only consider the first derivative of
A(x).
Taking the first derivative of ψ, we have:
 
dψ dA dφ iφ(x)
= + iA e
dx dx dx
So, the second derivative of ψ, which we will plug into Schrödinger’s equation, will be:
 2 !
d2 ψ d2 A dA dφ d2 φ dφ
= + 2i + iA 2 − A eiφ(x) (15)
dx2 dx2 dx dx dx dx

If we re-write Schrödinger’s equation such that d2 ψ/dx2 sits by itself, we can define the function
p(x) such that:
d2 ψ 2m(E − V (x)) p(x)2
= − ψ = − ψ (16)
dx2 ~2 ~2
Setting equation (15) equal to −p2 /~2 , we have:
 2
d2 A dA dφ d2 φ dφ p2
+ 2i + iA − A = − A
dx2 dx dx dx2 dx ~2
The above equation can be split into a real part:
"  2 #
d2 A p2 dφ
2
=A − 2 + (17)
dx ~ dx
2
Named after Gregor Wentzel, Hans Kramers, and Léon Brillouin.

6
and an imaginary part:
d2 φ
 
dA dφ d 2 dφ
2 +A 2 = A =0 (18)
dx dx dx dx dx
The solution to the imaginary part is just that A2 φ0 is equal to a constant, which we can call C 2 ,
meaning:
C
A= √ 0 (19)
φ
Recall that the WKB approximation requires us to set all derivatives of A of order-2 or higher
equal to zero. This means that the real part of Schrödinger’s equation reduces to:

p2
 
dφ dφ p(x)
= 2 ⇒ =±
dx ~ dx ~

Solving this simple differential equation for φ(x), we have:

Z
1
φ(x) = ± p(x)dx (20)
~
√ p
We know have everything we need to find the wavefunction. Note that φ0 = p(x)/~, but we
can just absorb ~ into the constant C. Thus, we have:
C 1
R
ψ(x) ≈ p e± ~ p(x)dx (21)
p(x)

Notice that I left the integral for φ(x) as an indefinite integral. This is fine because any integration
0
constant will appear in ψ as a eC , which can just be absorbed into C and ignored. However, since
p(x) isn’t strictly real (it depends on whether E > V (x) or E < V (x) at x), absorbing this constant
into C will make C a potentially imaginary number.
The general solution for ψ(x) will, of course, be a linear combination of the two solutions found
above:
C 1
R C0 1
R
ψ(x) ≈ p e ~ p(x)dx + p e− ~ p(x)dx (22)
p(x) p(x)
Notice, also, that the modulus-square of ψ(x) is given by:

|C|2
|ψ(x)|2 ≈ (23)
p(x)

so the probability of finding the particle at x is inversely proportional to the classical momentum
p at x. This should be obvious: if a particle is moving very fast, i.e. has a large p, then it won’t be
very likely to be measured in a location, since it moves on from a location very rapidly.
For tunneling, we want to consider the case where E is strictly less than V (x) throughout the
potential barrier, i.e. from 0 ≤ x ≤ a. This means that C will be imaginary, and everything must
be calculated with the modulus of p(x):

C 1 x
R D 1 x
R
ψ(x) ≈ p e ~ 0 |p(x)|dx + p e− ~ 0 |p(x)|dx (24)
|p(x)| |p(x)|

where I am using the coefficients in the same manner as in equation (6). Outside of the potential
barrier, we have the typical free particle solutions given by (2).

7
 Figure 3: Wavefunction vs. position for a tunneling particle.

While we could go through the same procedure as in the previous section, setting the wave-
functions and their derivatives equal at x = 0 and x = a, we can actually guess at the solution
based solely on looking at the amplitude throughout particle’s motion from −∞ to +∞. The
wavefunction vs. position is plotted in Figure 3.
As clear in the figure, the particle reaches the potential with an amplitude of A, which then
undergoes an expoential decay, becoming F after it exits the barrier. So, F should be related to A
like:
1 a
R
F ≈ Ae− ~ 0 |p(x)|dx
where this is an approximate solution not because we’re guessing as to what it is, but because the
amount of amplitude decay is approximate. So, the transmission coefficient, which is just |F |2 /|A|2 ,
is:
2 ap
 Z 
T ≈ exp − 2m[V (x) − E]dx (25)
~ 0
where I substituted |p(x)| into the solution as it was defined, but reversing the order of subtraction
E − V (x) → V (x) − E to make |p(x)| a real number.
Let’s compare the WKB result with our weak limit result. All of our WKB analysis was for a
general, potentially varying V (x). But the case of tunneling that we are considering, V (x) = V0 is
a constant. So, the WKB result is that:
r !
2m(V0 − E)
T ≈ exp −2 a
~2

where I pulled the ~ into the square root as ~2 to keep with the common convention. Looking back
to equation (13) for the weak limit solution, and all the way back to (6) for the definition of κ, the
transmission coefficient in the weak limit was found to be:
" r #
16E(E − V0 ) 2m(V0 − E)
T ≈ exp −2 a
V02 ~2

Clearly, both the WKB approximation and the weak limit give the same exponential decay of the
transmission coefficient, but the weak limit solution is going to be more accurate in the case of a
constant potential barrier, since it was derived as the asymptotic behavior of the exact solution.
However, the WKB approximation is going to, obviously, be the superior solution for a non-constant
potential barrier, as that’s what it was meant to be used for.

8
4 Gamow Theory of Alpha Decay
In Gamow’s theory of α-decay, he modeled the potential energy of the α-particle, over distance,
as being some constant, negative value −U within t the nucleus, extending from 0 to some R, and
then the potential energy being Coulomb for r > R. This is plotted in Figure 4 below.

Figure 4: Potential barrier for an α particle of energy E.

If the α-decay occurs in a nucleus with an atomic number Z + 2, then the Coulomb barrier is:

Ze2
V (r) = 2k (26)
r
Since this is clearly varying with position, we need to use our WKB result. If we define r2 to be
the value of r when it exits the Coulomb barrier, i.e. the point at which V (r2 ) = E, then r2 will
be given by:
Ze2
r2 = 2k (27)
E
Then, the transmission coefficient will be:
" Z s  #
2 r2 Ze2

T ≈ exp − 2m 2k − E dr
~ R r

Notice that we can say 2kZe2 = Er2 , so we can pull out a common factor of E from the square
root, giving: " √ #
2 2mE r2 r2
Z r
T ≈ exp − − 1dr
~ R r
The solution to this integral should just be looked up:
" √ r !#
2 2mE R
r2 cos−1
p
T ≈ exp − − R(r2 − R)
~ r2

For a realistic decay scenario, E should be fairly far below the Coulomb barrier at R, thus
r2  R. The arccosine term is dependent upon R/r2 , which tends to zero in the realistic case.

9
This means that θ → π/2, but remains slightly below it. If we define a very small angle  such that
θ = π/2 − , then we can use the angle-difference identity for cosine to show:
π  π  π 
cos θ = cos −  = cos cos  + sin sin  ≈ 
2 2 2
p p
since the small-angle approximation for sine is sin  ≈ . So, cos θ = R/r2 , and thus  = R/r2 ,
which means that the arccosine, which equals θ, is π/2 − , or:
r r
−1 R π R
cos = −
r2 2 r2
p √
So, noting that R(r2 − R) ≈ r2 R, the transmission coefficient is:
" √ #
2 2mE  π p 
T ≈ exp − r2 − 2 r2 R (28)
~ 2

It might make more sense to re-insert r2 as a function of Z and E, so that we don’t have to
worry about compute r2 for every case we come across. If we do so, we can define two constants,
E and R, such that:
√
2 π 2m
E = 2ke = 3.96 MeV1/2
~
√ (29)
√ m
R = 8 ke2 = 2.97 fm−1/2
~
such that the transmission coefficient becomes:
√
 
Z
T ≈ exp −E √ + R ZR (30)
E
We can use the semi-empirical mass formula (based on the liquid drop model of the atom, also
proposed by George Gamow) to define the atomic radius:

R = r0 A1/3 (31)

where r0 ≈ 1.25 fm, and A is the nucleon number of the radioactive nucleus. A should include the
4 nucleons of the α particle, since the boundary of the nucleus is determined prior to the α particle
beginning its tunneling, whereas we defined the proton number Z to be excluding the 2 protons in
the α particle, since we used it to define the Coulomb repulsion between the left-over nucleus and
the α particle. For instance, uranium-238, which is an α emitter, has 92 protons prior to decay.
For this problem, we would set A = 238, since the total number of nucleons including the α particle
defines R, but we would set Z = 90 to define the Coulomb barrier, since it is due to the interaction
between the left-over nucleus (thorium-234) and the α particle.
The last thing to compute is the lifetime of an α emitting nucleus. If we consider an α particle
as bouncing around on the inside of an atom, then it has to traverse, on average, the diameter
of the atom between collisions, or 2R. If it moves with a speed of v inside the nucleus, then the
frequency of collisions is:
v
f=
2R

10
Every time the α particle hits the Coulomb barrier, i.e. the boundary of the nucleus, it has a
probability T of penetrating the barrier and escaping. So, the probability of emission will be
vT /2R, and the lifetime τ will the inverse of this probability:
√
 
v Z
τ≈ exp −E √ + R ZR (32)
2R E
There are many ways of estimating the speed v of an α particle in a nucleus. The simplest way is
to use a non-relativistic approach and set E = 21 mv 2 . The actual prediction of τ isn’t going to be
great, because there isn’t really any particularly
√ good way of estimating v. However, the behavior
of ln τ is dominated by the factor of 1/ E, and this behavior agrees perfectly with experiment, as
shown in Figure 53 .

Figure 5: Theoretic vs. observed half-lives of α emitters. Image credit: Griffiths.

3
Taken from D.J. Griffiths, “Introduction to Quantum Mechanics,” Prentice Hall (1995).

11