Where Does the Transmission Coefficient of Quantum Tunneling Come From?

— Following the derivation in lecture notes (6)–(9) (for second- and third-year undergraduates)

The simulator draws the transmission coefficient from the formula T=[1+V02sinh2⁡bL4E(V0−E)]−1 — a result that comes from nothing more than a system of simultaneous linear equations from four boundary conditions. No difficult mathematics is needed. We follow the flow of the notes and trace every step without skipping any.

The conclusion first The wave function does not become 0 even inside the barrier. For E<V0 the solution is simply a real exponential e±bx (a superposition of a decaying term and a growing term), and the square of the small amplitude that remains on the other side is itself the transmission coefficient. For a thick barrier, T≃e−2bL — and because it acts in the exponent, even a slightly thicker barrier changes it by orders of magnitude.

0. Setup

This is the setup of lecture notes (6). In the path of a particle of mass m moving along the x axis, we place a box-shaped potential of height V0 and thickness L.

V(x)= { 0(x < 0 or L < x) V0(0 ≤ x ≤ L)

Classically, a particle with E<V0 is always reflected. To see what happens in quantum theory, we solve the time-independent Schrödinger equation −ℏ22md2ψdx2+Vψ=Eψ region by region.

1. Intuition First (the 30-Second Version)

The wave function satisfies ψ″=(2m/ℏ2)(V−E)ψ — and the sign of the right-hand side alone determines what the solution looks like.

If the barrier has only a finite thickness, the wave gets out before it has finished decaying. The square of the amplitude where it emerges is the transmission coefficient. The remark in the margin of the notes, “This is essentially the same as the seep-out effect!” refers to exactly this.

2. Solving Region by Region and Matching — Lecture Notes (6)

2.1 General solutions in the three regions

Let k=2mE/ℏ, b=2m(V0−E)/ℏ.

(1) ψ(x)= { Aeikx+Be−ikx (x < 0: incident and reflected waves) Debx+Fe−bx (0 ≤ x ≤ L: real exponentials) Ceikx (L < x: transmitted wave only)
Why not put a left-moving wave on the right-hand side? Because there is nothing left in the region x>L that could reflect the wave. The very setup of “a particle is fired in from the left” is built into this one line. Conversely, if we kept both the right-moving and left-moving waves, we would be solving a different physical problem (an experiment with particles fired from both sides).

2.2 There are four matching conditions

Since V has only finite jumps, ψ and ψ′ must be continuous at x=0 and x=L.

(2) A+B =D+F ik(A−B) =b(D−F) CeikL =DebL+Fe−bL ikCeikL =b(DebL−Fe−bL)

2.3 Solving — from back to front

There are five unknowns, A,B,C,D,F and only four equations. The overall size is not fixed, so setting C=1 and working from the back to the front spares us from solving the simultaneous equations. First, adding and subtracting the 3rd and 4th equations of Eq. (2) gives

(3) D=b+ik2be(ik−b)LC , F=b−ik2be(ik+b)LC

In the same way, adding and subtracting the 1st and 2nd equations lets us write A,B in terms of D,F, and substituting Eq. (3) gives

(4) AC= eikL [2kbcosh⁡bL −i(k2−b2) sinh⁡bL] 2kb , BC= −i(k2+b2) eikLsinh⁡bL 2kb
The notes write the same quantity in the form B/A=(k2+b2)(ebL−e−bL)/[⋯] — the only difference is that here ebL±e−bL has been rewritten as 2cosh / 2sinh; it is the same equation.

2.4 Transmission and reflection coefficients

The transmission and reflection coefficients are T=|C/A|2, R=|B/A|2. Taking the absolute value of Eq. (4) and using cosh2=1+sinh2 and 4k2b2+(k2−b2)2=(k2+b2)2, the result collapses into a surprisingly neat form.

(5) T= 4k2b2 (k2+b2)2sinh2⁡bL+4k2b2 , R= (k2+b2)2sinh2⁡bL (k2+b2)2sinh2⁡bL+4k2b2

The denominators are common, and the numerators are exactly the two terms of the denominator. Hence T+R=1 is obvious at a glance — this is what the remark in the margin of the notes, “Indeed, T+R=1,” refers to. It is the best possible evidence that the answer is correct: probability is conserved.

Why is the square of the amplitude ratio enough? The transmission coefficient is, strictly speaking, defined as the ratio of probability currents (current of the transmitted wave ÷ current of the incident wave). Here the potential is the same on both sides of the barrier, so the wavenumber k, that is, the speed at which the wave travels, is common to both sides, and the ratio of the currents is simply the square of the amplitude ratio. For a step with different potential heights on the two sides, the ratio of the wavenumbers on the transmitted and incident sides enters as a factor, so this coincidence is not a matter of course.

Finally, we substitute k2=2mE/ℏ2, b2=2m(V0−E)/ℏ2. Here it helps that k2+b2=2mV0/ℏ2 contains no E, and we obtain

(6) T=[1+ V02sinh2⁡bL 4E(V0−E) ]−1 , R=1−T

This is the formula that tab ① of the simulator draws.

3. When the Energy Is Higher Than the Barrier — Lecture Notes (7)

For E>V0 we have b2<0. There is no need to solve the problem again. Just substitute k′2=2m(E−V0)/ℏ2=−b2 — that is, b=ik′ — directly into Eq. (6).

(7) sinh⁡bL= sinh⁡ik′L= eik′L−e−ik′L2 =isin⁡k′L ⟹ sinh2⁡bL= −sin2⁡k′L

Together with V0−E=−(E−V0), the two minus signs cancel, giving

(8) T=[1+ V02sin2⁡k′L 4E(E−V0) ]−1
This is the biggest difference from classical theory Classically, if E>V0, the particle passes through 100 % of the time. But Eq. (8) generally gives T<1 — the particle is reflected even though it has enough energy to get over the hill. Conversely, only when sin⁡k′L=0, that is, k′L=nπ, does the transmission coefficient return to T=1 (resonant transmission). This is the condition under which the barrier thickness is an integer multiple of the half wavelength inside the barrier, so that the waves reflected from the two boundaries exactly cancel each other. In optics, it is the same as a layer of half-wavelength thickness becoming invisible with respect to reflection (Fabry–Pérot resonance). The anti-reflection coating that is often cited uses a different condition, a thickness of a quarter wavelength, so the thickness conditions do not coincide.

The energies at which resonance occurs are E=V0+n2π2ℏ2/2mL2. For the simulator’s default values (V0=10 eV, L=1 nm, electron), they are 10.376, 11.504, and 13.384 eV. You can see these peaks in the T–E curve of tab ①.

4. The Thick-Barrier Approximation and the Gamow Factor — Lecture Notes (8)

4.1 Replacing sinh with an exponential

If bL≫1, then sinh⁡bL=(ebL−e−bL)/2≃ebL/2. Furthermore, the fraction in Eq. (6) is much larger than 1, so the 1 inside the brackets can be dropped as well, giving

(9) T≃ 16E(V0−E)V02 e−2bL ≃ e−2bL =exp⁡[−2Lℏ2m(V0−E)]

In the last equality, the notes drop even the prefactor. This is what the sentence “since the value of T is determined by the exponent” refers to. Let us check numerically how crude this is.

bL T (exact, Eq. (6)) With prefactor, Eq. (9) Error e−2bL
13.78×10−14.55×10−1+20 %1.35×10−1
26.00×10−26.15×10−2+2.5 %1.83×10−2
38.30×10−38.33×10−3+0.3 %2.48×10−3
51.53×10−41.53×10−4+0.006 %4.54×10−5
8.876.59×10−86.59×10−8< 10−61.96×10−8

Eq. (9), which keeps the prefactor, is accurate to within 0.3 % for bL≥3. On the other hand, the bare e−2bL in the rightmost column stays off by a factor of about 3 — this is because the prefactor 16E(V0−E)/V02 is a constant (3.36 in this example) and does not go away however large bL becomes. The right attitude is to accept that this is an approximation to be used only when “getting the order of magnitude right is enough.” This is why the simulator lists both.

A real example of how orders of magnitude matter — the scanning tunneling microscope Regard the gap between the metal tip and the sample as L, and let the barrier height minus the electron energy, V0−E, equal the work function of the metal (about 4 eV); then b≃10 nm−1. The tunneling current is proportional to e−2bL and therefore increases eightfold — almost an order of magnitude — when the gap shrinks by only 0.1 nm. The scanning tunneling microscope can distinguish bumps the size of a single atom thanks to the fact that the dependence sits in the exponent.

4.2 Toward a hill-shaped potential — the Gamow factor

The second half of lecture notes (8) is the clever part. Even if the barrier is not box-shaped, we can regard it as a stack of thin boxes and simply multiply their transmission coefficients. Since multiplication becomes addition in the exponent,

(10) T≃∏j exp⁡[−2Δxℏ2m(V(xj)−E)] ⟶Δx→0 exp⁡[−2ℏ∫x1x22m(V(x)−E)dx]

This is the Gamow factor (it agrees with the result of the WKB approximation). The integration runs only over the part where V(x)>E, that is, the classically forbidden region.

⚠ This “multiplication” is an intuition, not an exact scattering calculation When thin boxes are lined up, a reflected wave arises at every boundary and their phases interfere, so in reality T≠∏jTj — to do it exactly, one would multiply not the transmission coefficients but the transfer matrices (scattering matrices). What is added up here is only the exponent of the factor exp⁡[−Δxℏ2m(V(x)−E)] by which the amplitude decreases in each interval. The phases of the reflections at the boundaries and the prefactor (the counterpart of 16E(V0−E)/V02 in Section 4.1) are discarded in this argument. Even so, the exponent agrees with the WKB answer obtained semiclassically, and it is the exponent that determines the order of magnitude.

Left: a plot of |ψ|2 around the barrier. It decays exponentially inside, and only a small amplitude remains on the right (the vertical axis is logarithmic). Right: energy dependence of T. The peaks that appear for E>V0 correspond to resonant transmission (k′L=nπ).

5. Application: α Decay — Lecture Notes (9)

This is the most famous use of the Gamow factor. Inside the nucleus the α particle is confined by the nuclear force, and outside it feels the Coulomb repulsion. Below, M is the mass of the α particle, Z is the atomic number of the daughter nucleus that remains after the α particle has left, R is the radius of the nucleus (a different symbol from the R that we have used so far for the reflection coefficient). The denominator of the Coulomb term is 2π rather than 4π because the charge 2e of the α particle has been put into the numerator.

V(r)= { −V0(r < R) 12πε0Ze2r(r > R)

The classical turning point r0 is determined by V(r0)=E. Using this, the expression under the square root takes the simple form E(r0/r−1) — and with the substitution r=r0sin2θ the integral can be evaluated.

(11) ∫Rr0 r0r−1dr =r0[ π2− sin−1Rr0 −Rr0(1−Rr0) ]
⚠ Be careful how the approximations are matched here The notes assume r0≫R and drop the second term by writing sin−1R/r0≃0 — but they keep the third term alone, as R/r0. However, since sin−1u≃u, these two terms are of the same order. If both are kept consistently, the coefficient doubles.
T≃exp⁡[ − 2MEℏ r0(π−4Rr0) ]
Because this sits in the exponent, the difference has a direct effect. In actual α decay, where R/r0≃0.1–0.2, the form in the notes overestimates the integral by 32–59 %, whereas the form above gives only a 1–4 % underestimate. This makes a good exercise in being consistent about the order to which an expansion is kept.

6. What Does the Simulator Compute?

Tab ① evaluates Eqs. (6) and (8) above directly. Tab ② is more naive: it simply solves the time-dependent Schrödinger equation numerically and fires a wave packet at the barrier.

What it doesDetails
Internal unitsnm / eV / fs (ℏ=0.6582 eV·fs, ℏ2/2m=0.0381 eV·nm2). The numbers stay of order 1, so nothing overflows
Time evolutionSplit-operator method (split-step Fourier method). The kinetic energy is applied in wavenumber space and the potential in real space
EdgesAbsorbing layers are placed there, and the probability that escapes is accumulated separately for the left and right sides
What it is compared withThe average over the momentum distribution of the wave packet: ⟨T⟩
A wave packet and a single energy do not agree — and this is not an error Eq. (6) is the answer for a single energy. A wave packet, however, always has a spread in energy (the price of localizing the position). For this reason the transmission probability actually measured should be compared not with T(E), but with the average ⟨T⟩ weighted by the momentum distribution. Because T varies exponentially with E, the high-energy side of the spread dominates overwhelmingly, and the average comes out larger than the value at the center (this applies to the tunneling regime E<V0. For E>V0, T oscillates because of resonances, so at the top of a resonance peak the average is, conversely, smaller). The wider the spatial width of the wave packet — that is, the narrower its spread in energy — the closer the two become.

7. Checking Numerically

Since the derivation was done entirely by hand, it has been checked independently by numerical computation. The table compares the complex amplitudes obtained by solving the four equations of Eq. (2) directly, starting from C=1, with the closed forms in the notes.

What was checkedAgreement
|B/A|2 from solving the boundary conditions directly vs Eq. (4) (three combinations of k, b, L) relative difference < 1×10−15
Likewise, |C/A|2 vs Eq. (5) relative difference < 1×10−15
T+R=1exact (< 10−16)
Rewriting Eq. (5) into Eq. (6)relative difference < 5×10−16
Analytic continuation b=ik′ in Eq. (7) → Eq. (8)exact agreement
Resonance k′L=nπ gives T=1 12 digits
Closed form of Eq. (11) vs numerical integrationrelative difference < 2×10−11
On the differences from the numbers in the notes The notes give bL≃8.9, T≃6.25×10−8, and R≃0.136. Here, bL and T are from the tunneling example with E=7 eV, whereas R, back-calculated from the numbers, corresponds to E≃13 eV — that is, it appears to come from another example with E>V0, to which Eq. (8) applies (both with V0=10 eV, L=1 nm). For the tunneling example, R=1−T≃1, so there is no point in listing it here. With the CODATA values we get bL=8.874, T=6.59×10−8, and R=0.149. The difference comes only from the rounding ℏ≃1.05×10−34 — which in terms of bL is a difference of 0.4 %. That 0.4 % is amplified through T∝e−2bL to 5 %, and in R it is amplified to 10 % through the slope of sin. The lesson: carry one extra significant figure for anything that sits in an exponent or inside a trigonometric function.

8. Summary

  1. Solve the Schrödinger equation in each region and join the solutions with the four continuity conditions on ψ and ψ′. That is all — Eqs. (1) and (2).
  2. Setting C=1 and working from the back to the front yields A,B without solving the simultaneous equations — Eqs. (3) and (4).
  3. Taking absolute values gives a common denominator, so that T+R=1 is obvious at a glance — Eqs. (5) and (6).
  4. For E>V0 we simply substitute b=ik′. sinh turns into sin, and resonant transmission appears — Eqs. (7) and (8).
  5. For a thick barrier, T≃e−2bL. Cutting the barrier into thin slices and multiplying yields the Gamow factor, which can be used for a barrier of any shape that varies slowly (strictly, one should multiply transfer matrices, but the exponent is determined this way) — Eqs. (9) and (10).
  6. Applying it to the Coulomb barrier gives the lifetime for α decay — Eq. (11). A single formula works all the way from electrons in the laboratory to atomic nuclei.

Mochizuki Research Group, Tokyo University of Science / from lecture notes (6) “Quantum Tunneling”; (7) “The Box-Shaped Potential in the Case of Particle Energy E > Potential V0”; (8) “Approximate Expression for the Transmission Coefficient and General Hill-Shaped Potentials”; and (9) “Theory of α Decay.” The numerical checks use direct calculation of the complex amplitudes and Simpson’s rule. The equations are typeset in the browser’s native MathML (no external libraries or network communication are used at all).