Skip to content
LumenaryAcademy
Functional 75 min

Electronic structure and bands

Why copper conducts, diamond insulates and silicon sits in between — from free electrons and atomic levels to the bands of a crystal, computed.

Code tested · 6AI-draftedAwaiting review by a materials scientist

Your guides: Chloe · The Materials Scientist, Margaret · The Tester, Gideon · The Failure Analyst

MotivationH

Twenty-four orders of magnitude

Measure how strongly different solids resist an electric current and the numbers refuse to stay in any sensible range. Copper has a resistivity of about 1.7×10−81.7 \times 10^{-8} Ω m. Pure silicon at room temperature, about 2×1032 \times 10^{3} Ω m. Fused silica glass, more than 101610^{16} Ω m. No other everyday property of materials varies by twenty-four orders of magnitude. Density varies by a factor of about forty between lithium and osmium; stiffness by about five orders of magnitude between soft rubber and diamond.

Stranger still, the order is not what the bonding lesson might lead you to expect. Every one of these solids has plenty of electrons: a cubic centimetre of diamond holds about 102410^{24} of them. And it is not simply that metals have "loose" electrons: aluminium has three valence electrons per atom and conducts superbly; silicon has four and barely conducts at all; carbon, also with four, in the form of diamond is one of the best insulators known, while in the form of graphite it conducts.

The answer, worked out between 1928 and 1931 by Arnold Sommerfeld, Felix Bloch, Rudolf Peierls, Alan Wilson and others, is one of the great successes of quantum mechanics. Electrons in a crystal can only have energies in certain ranges, called bands, separated by forbidden gaps, and whether a solid conducts depends on whether its highest occupied band is completely full. This lesson builds the idea in three ways, each computed: free electrons in a box, atoms brought together until their levels merge, and waves moving through a periodic potential. Then it uses bands to explain metals, semiconductors and insulators, and introduces the two quantities engineers use every day, the effective mass and the density of states.

You will need the quantum mechanics of quantum mechanics for chemists (the particle in a box, Pauli's principle) and the molecular orbitals of molecular orbitals, whose Hückel model is where the tight-binding idea starts. The reciprocal lattice from X-ray diffraction returns here as the home of electron wavevectors.

IntuitionH

Free electrons, and where they fail

The simplest model of a metal ignores the ions altogether. The valence electrons are treated as a gas of independent particles in a box the size of the crystal. Each is a standing (or travelling) wave with wavevector k\mathbf k and energy

E=ℏ2k22me.E = \frac{\hbar^2 k^2}{2m_e}.

With periodic boundary conditions on a cube of side LL, the allowed wavevectors form a fine cubic grid with spacing 2π/L2\pi/L, so each state occupies a volume (2π/L)3(2\pi/L)^3 of k\mathbf k-space. Pauli's principle allows two electrons (spin up and down) per k\mathbf k. At absolute zero, NN electrons fill the lowest states: every k\mathbf k inside a sphere, the Fermi sphere, of radius kFk_F. Counting states,

N=2⋅43πkF3(2π/L)3⟹kF=(3π2n)1/3,EF=ℏ2kF22me,N = 2 \cdot \frac{\tfrac43\pi k_F^3}{(2\pi/L)^3} \quad\Longrightarrow\quad k_F = (3\pi^2 n)^{1/3}, \qquad E_F = \frac{\hbar^2 k_F^2}{2m_e},

where n=N/L3n = N/L^3 is the electron density. The highest filled energy, EFE_F, is the Fermi energy. The program computes it for three metals from their crystal structures, and checks the state-counting argument by brute force.

python
import numpy as np

hbar = 1.054571817e-34   # J s
me = 9.1093837015e-31    # kg
eV = 1.602176634e-19     # J
kB = 1.380649e-23        # J/K

# Free-electron densities from the crystal structure:
# (lattice parameter in nm, atoms per cubic cell, valence electrons per atom), room temperature
metals = {"Na": (0.4291, 2, 1), "Cu": (0.3615, 4, 1), "Al": (0.4049, 4, 3)}
print("metal  n/(1e28 m^-3)  E_F/eV  T_F/K     v_F/(1e6 m/s)  wavelength/nm")
for name, (a, atoms, z) in metals.items():
    n = atoms * z / (a * 1e-9) ** 3
    kF = (3 * np.pi**2 * n) ** (1 / 3)
    EF = hbar**2 * kF**2 / (2 * me)
    print(f"  {name}     {n / 1e28:6.2f}       {EF / eV:5.2f}  {EF / kB:7.0f}   {hbar * kF / me / 1e6:6.2f}"
          f"          {2 * np.pi / kF * 1e9:.3f}")

# How many electrons are within kT of the Fermi level at room temperature (copper)?
n = 4 / (0.3615e-9) ** 3
EF = hbar**2 * (3 * np.pi**2 * n) ** (2 / 3) / (2 * me)
kT = kB * 300
print(f"\ncopper at 300 K: kT = {kT / eV * 1000:.1f} meV, kT/E_F = {kT / EF:.4f}")
print(f"  fraction of electrons in a window kT wide at E_F: {1.5 * kT / EF:.4f}")

# Check the density of states by brute force: count the standing-wave states of a cube
# (periodic boundaries) whose energy is below E, and compare with the formula N(E) = V k^3 / (3 pi^2).
L_states = 40                          # count integer triples (nx, ny, nz) in a sphere
r = np.arange(-L_states, L_states + 1)
nx, ny, nz = np.meshgrid(r, r, r, indexing="ij")
n2 = (nx**2 + ny**2 + nz**2).ravel()
print("\n (k L / 2 pi)^2   states counted   formula   ratio")
for R in [5, 10, 20, 40]:
    counted = 2 * np.count_nonzero(n2 <= R**2)       # x2 for spin
    formula = 2 * 4 / 3 * np.pi * R**3
    print(f"   {R**2:6d}        {counted:9d}   {formula:9.0f}   {counted / formula:.4f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The Fermi energies are several electronvolts: 3.1 eV for sodium, 7.0 eV for copper, 11.7 eV for aluminium. Expressed as temperatures, TF=EF/kBT_F = E_F/k_B, they are tens of thousands of kelvin. The electrons at the top of the Fermi sea move at over a million metres per second even at absolute zero, purely because Pauli's principle forbids them all from sitting still. And at room temperature kBTk_BT is only 0.4 per cent of copper's Fermi energy: only about half a per cent of the electrons, those within a few kBTk_BT of EFE_F, can be thermally excited into empty states. That single fact solved a puzzle that had sunk the classical theory of Paul Drude (1900): if every free electron absorbed heat like a gas molecule, metals would have a much larger heat capacity than they do.

The brute-force count also confirms the step from a grid of states to a smooth formula: the number of states with energy below EE grows as k3∝E3/2k^3 \propto E^{3/2}, so the number per unit energy, the density of states, grows as E\sqrt E. The agreement improves as the sphere grows, because a real crystal has about 102310^{23} states and the grid becomes effectively continuous.

IntuitionH

From atomic levels to bands

Come at the problem from the opposite end. Start with isolated atoms, each with sharp energy levels, and bring them together.

Two hydrogen atoms make a molecule whose 1s orbitals combine into a bonding and an antibonding orbital, split by twice the coupling between them (molecular orbitals). Six carbon pp orbitals in a ring make benzene's six π orbitals. In the Hückel model each atom contributes a level at energy ε0\varepsilon_0, and each pair of neighbours is coupled by a hopping energy −t-t (the chemist's β\beta). For a ring of NN atoms the Hamiltonian is a matrix with ε0\varepsilon_0 on the diagonal and −t-t between neighbours, and its eigenvalues are

Ej=ε0−2tcos⁡ ⁣(2πjN),j=0,1,…,N−1.E_j = \varepsilon_0 - 2t\cos\!\left(\frac{2\pi j}{N}\right), \qquad j = 0, 1, \ldots, N-1.

Whatever NN is, the levels lie between ε0−2t\varepsilon_0 - 2t and ε0+2t\varepsilon_0 + 2t. A crystal is the case N≈1023N \approx 10^{23}: the levels are packed so closely that they form a continuous band of width 4t4t. This is the tight-binding picture, and it is the same physics as the bonding and antibonding orbitals of chemistry, carried to its limit. The program builds the matrices and diagonalises them.

python
import numpy as np

def ring(N, t=1.0, eps0=0.0):
    """Tight-binding Hamiltonian of N identical atoms in a ring: site energy eps0, hopping -t."""
    H = eps0 * np.eye(N)
    for i in range(N):
        j = (i + 1) % N
        H[i, j] = H[j, i] = -t
    return H

t = 1.0                                   # hopping energy in eV
print("two atoms (a molecule):", np.linalg.eigvalsh(np.array([[0, -t], [-t, 0]])), "eV")
for N in [6, 12, 1000]:
    E = np.linalg.eigvalsh(ring(N, t))
    k = 2 * np.pi * np.arange(N) / N       # k a for the allowed wavevectors
    formula = np.sort(-2 * t * np.cos(k))
    print(f"ring of {N:4}: lowest {E[0]:+.3f}, highest {E[-1]:+.3f} eV, "
          f"distinct levels {len(np.unique(E.round(9)))}, max |E - (-2t cos ka)| = {np.abs(E - formula).max():.1e}")

# Density of states of the long ring: how many levels fall in each energy window
E = np.linalg.eigvalsh(ring(1000, t))
counts, edges = np.histogram(np.clip(E, -2, 2), bins=10, range=(-2, 2))
print("\nlevels per 0.4 eV window (1000-atom ring):")
for c, lo in zip(counts, edges):
    print(f"  {lo:+.1f} to {lo + 0.4:+.1f} eV  {c:4d} " + "#" * (c // 5))

# Two kinds of atom in each cell (site energies +D and -D): a gap opens
def diatomic(Ncells, t=1.0, D=0.5):
    N = 2 * Ncells
    H = ring(N, t)
    H[np.arange(N), np.arange(N)] = np.where(np.arange(N) % 2 == 0, D, -D)
    return H

for D in [0.0, 0.25, 0.5, 1.0]:
    E = np.linalg.eigvalsh(diatomic(500, t, D))
    gap = E[500] - E[499]                  # between the lower and upper band
    print(f"site energies +-{D:.2f} eV: lower band {E[0]:+.2f} to {E[499]:+.2f}, "
          f"upper band {E[500]:+.2f} to {E[-1]:+.2f}, gap {gap:.3f} eV")

# Effective mass at the bottom of the band, from the curvature of E(k) = -2t cos(ka)
hbar2_over_2me = 0.0380998                 # hbar^2 / (2 m_e) in eV nm^2
a = 0.25                                   # nm
for t in [2.0, 1.0, 0.25, 0.05]:
    dk = 1e-3
    k = np.array([-dk, 0.0, dk])
    Ek = -2 * t * np.cos(k * a)
    curv = (Ek[0] - 2 * Ek[1] + Ek[2]) / dk**2         # d2E/dk2 in eV nm^2
    m_ratio = 2 * hbar2_over_2me / curv                  # m*/m_e = hbar^2 / (m_e d2E/dk2)
    print(f"t = {t:4.2f} eV: band width {4 * t:4.1f} eV, m*/m_e = {m_ratio:6.2f} "
          f"(formula hbar^2/(2 t a^2 m_e) = {hbar2_over_2me / (t * a**2):.2f})")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Four lessons are in the output.

  1. N levels in, N levels out. Two atoms give two levels at ±t\pm t; a thousand give a thousand levels, all inside [−2t,2t][-2t, 2t]. The eigenvalues match −2tcos⁡(ka)-2t\cos(ka) to rounding error, where k=2πj/(Na)k = 2\pi j/(Na) plays the role of a wavevector. Each band therefore holds exactly NN orbital states, or 2N2N electrons counting spin. That counting rule decides which solids are metals.
  2. The density of states is not flat. The levels crowd together at the band edges, where E(k)E(k) is flat, and thin out in the middle, where it is steep. In one dimension the density of states actually diverges at the edges (a van Hove singularity).
  3. Two atoms per cell open a gap. When neighbouring sites have different energies, +Δ+\Delta and −Δ-\Delta (think of the cation and anion of an ionic solid), the single band splits into two with a gap of exactly 2Δ2\Delta between them. A different alternation, of the hopping rather than the site energy, does the same to the conjugated polymer polyacetylene, whose single and double bonds alternate; doping that polymer turns it into a conductor, work recognised by the 2000 Nobel Prize in Chemistry to Hideki Shirakawa, Alan MacDiarmid and Alan Heeger.
  4. Narrow bands mean heavy electrons. Near the bottom of the band E≈ε0−2t+ta2k2E \approx \varepsilon_0 - 2t + t a^2 k^2, which looks like a free electron with a different mass, m∗=ℏ2/(2ta2)m^* = \hbar^2/(2ta^2). Weak hopping, a narrow band, gives a heavy effective mass: with a=0.25a = 0.25 nm, a 4 eV wide band behaves like electrons of mass 0.61 me0.61\,m_e and a 0.2 eV band like electrons twelve times heavier than free ones.
1 atom2 atoms6 atomscrystalspgaps bandp bandenergy
How atomic levels become bands. Each atom has a lower (s-like) and a higher (p-like) level. Two atoms split each level in two; a ring of six gives four distinct levels from each; a crystal of very many atoms gives continuous bands of width four times the hopping energy. The higher orbitals overlap more, so their band is wider. If the bands do not reach each other, a gap of forbidden energies remains between them.
DefinitionsH

Bloch waves, bands and the Fermi level

Both pictures are limits of one theory. The electrons of a crystal move in a potential with the periodicity of the lattice, and that symmetry alone fixes the form of their wavefunctions.

Theorem— Bloch's theorem

Let V(r+R)=V(r)V(\mathbf r + \mathbf R) = V(\mathbf r) for every lattice vector R\mathbf R. Then the eigenstates of H=−ℏ22m∇2+VH = -\frac{\hbar^2}{2m}\nabla^2 + V can be chosen in the form

ψnk(r)=eik⋅r unk(r),unk(r+R)=unk(r),\psi_{n\mathbf k}(\mathbf r) = e^{i\mathbf k\cdot\mathbf r}\,u_{n\mathbf k}(\mathbf r), \qquad u_{n\mathbf k}(\mathbf r + \mathbf R) = u_{n\mathbf k}(\mathbf r),

equivalently ψnk(r+R)=eik⋅R ψnk(r)\psi_{n\mathbf k}(\mathbf r + \mathbf R) = e^{i\mathbf k\cdot\mathbf R}\,\psi_{n\mathbf k}(\mathbf r): a plane wave modulated by a function with the period of the lattice.

Proof

Let TRT_{\mathbf R} shift a function by R\mathbf R, (TRψ)(r)=ψ(r+R)(T_{\mathbf R}\psi)(\mathbf r) = \psi(\mathbf r + \mathbf R). Because VV is periodic and the Laplacian does not care where the origin is, TRT_{\mathbf R} commutes with HH; the translations also commute with each other. Commuting operators have common eigenfunctions, so choose ψ\psi to be an eigenfunction of every TRT_{\mathbf R} with eigenvalue c(R)c(\mathbf R). Shifting twice is shifting by the sum, so c(R1+R2)=c(R1) c(R2)c(\mathbf R_1 + \mathbf R_2) = c(\mathbf R_1)\,c(\mathbf R_2), and a normalisable state in a large crystal with periodic boundaries needs ∣c∣=1|c| = 1. The only such functions are c(R)=eik⋅Rc(\mathbf R) = e^{i\mathbf k\cdot\mathbf R} for some vector k\mathbf k. Then u(r)=e−ik⋅rψ(r)u(\mathbf r) = e^{-i\mathbf k\cdot\mathbf r}\psi(\mathbf r) satisfies u(r+R)=e−ik⋅re−ik⋅R eik⋅Rψ(r)=u(r)u(\mathbf r + \mathbf R) = e^{-i\mathbf k\cdot\mathbf r}e^{-i\mathbf k\cdot\mathbf R}\,e^{i\mathbf k\cdot\mathbf R}\psi(\mathbf r) = u(\mathbf r).

∎

Definition— crystal momentum, Brillouin zone

The vector ℏk\hbar\mathbf k is the crystal momentum. It is defined only up to a reciprocal lattice vector G\mathbf G (the vectors of X-ray diffraction, with eiG⋅R=1e^{i\mathbf G\cdot\mathbf R} = 1), since k\mathbf k and k+G\mathbf k + \mathbf G give the same phases. Every distinct k\mathbf k can therefore be drawn inside one cell of the reciprocal lattice, the first Brillouin zone: in one dimension with period aa, the interval −π/a<k≤π/a-\pi/a < k \le \pi/a.

Definition— band, band gap, band structure

For each k\mathbf k the Schrödinger equation has a ladder of solutions, labelled by the band index nn, with energies En(k)E_n(\mathbf k). As k\mathbf k runs over the Brillouin zone each En(k)E_n(\mathbf k) sweeps out an energy range, a band. Energies covered by no band form a band gap. The set of curves En(k)E_n(\mathbf k) along lines through the zone is the band structure.

Definition— Fermi–Dirac distribution, Fermi level

In thermal equilibrium at temperature TT, a state of energy EE is occupied with probability

f(E)=1e(E−μ)/kBT+1.f(E) = \frac{1}{e^{(E-\mu)/k_BT} + 1}.

The energy μ\mu at which f=12f = \tfrac12 is the chemical potential of the electrons, which in materials science and device physics is called the Fermi level, EFE_F. In a metal at T=0T = 0 it equals the Fermi energy of the free-electron model; in a semiconductor or insulator it usually lies inside the gap, where there are no states at all.

Definition— effective mass

Near a band minimum or maximum, E(k)E(\mathbf k) is approximately quadratic, and an electron there responds to forces as if it had the effective mass

1m∗=1ℏ2d2Edk2\frac{1}{m^*} = \frac{1}{\hbar^2}\frac{d^2E}{dk^2}

(in three dimensions, a tensor of second derivatives). Its group velocity is v=1ℏdEdkv = \frac{1}{\hbar}\frac{dE}{dk}.

Definition— density of states

The density of states g(E)g(E) is the number of electron states per unit energy (per unit volume, or per atom): g(E) dEg(E)\,dE counts the states with energies between EE and E+dEE + dE. For free electrons in three dimensions, g(E)=12π2(2mℏ2)3/2Eg(E) = \frac{1}{2\pi^2}\left(\frac{2m}{\hbar^2}\right)^{3/2}\sqrt{E} per unit volume, counting both spins.

The wave picture explains why gaps appear. An electron wave whose wavelength matches the lattice, k=π/ak = \pi/a, is Bragg-reflected exactly as an X-ray is. Its forward and backward waves combine into two standing waves: one piles charge onto the ions, where the potential energy is low, and one piles it between them, where it is high. Same wavelength, two different energies: the difference is the gap.

ComputationC

The Kronig–Penney crystal, solved numerically

In 1931 Ralph Kronig and William Penney found a crystal simple enough to solve exactly: a one-dimensional row of square potential wells of width ww, separated by barriers of height V0V_0 and width bb, with period a=w+ba = w + b. Inside a well the solution is a combination of e±iαxe^{\pm i\alpha x} with α=2mE/ℏ\alpha = \sqrt{2mE}/\hbar; inside a barrier, of e±βxe^{\pm\beta x} with β=2m(V0−E)/ℏ\beta = \sqrt{2m(V_0 - E)}/\hbar. Requiring ψ\psi and ψ′\psi' to be continuous at the walls, and applying Bloch's theorem across one period, leaves a single condition (for E<V0E < V_0):

cos⁡ka=cos⁡αw cosh⁡βb+β2−α22αβ sin⁡αw sinh⁡βb.\cos ka = \cos\alpha w\,\cosh\beta b + \frac{\beta^2 - \alpha^2}{2\alpha\beta}\,\sin\alpha w\,\sinh\beta b.

Above the barrier, β\beta becomes imaginary and the hyperbolic functions turn into ordinary ones. The left side lies between −1-1 and 11. The right side does not have to: wherever it strays outside that range, no real kk exists, so no Bloch wave exists, so that energy is forbidden. The bands can be found by scanning the energy and checking the size of the right-hand side, which is exactly what the program does. It then recomputes the band edges by an entirely different method, writing ψ\psi as a sum of 121 plane waves ei(k+G)xe^{i(k + G)x} and diagonalising the resulting Hamiltonian matrix, as electronic-structure codes do for real crystals.

python
import numpy as np

C = 0.0380998               # hbar^2 / (2 m_e) in eV nm^2
a, b, V0 = 0.5, 0.1, 3.0    # period (nm), barrier width (nm), barrier height (eV)
w = a - b                   # well width (nm)

def kp(E):
    """Right-hand side of the Kronig-Penney condition cos(k a) = kp(E), for an array of energies."""
    E = np.asarray(E, dtype=float)
    al = np.sqrt(E / C)
    below = E < V0
    be = np.sqrt(np.abs(V0 - E) / C)                    # decay rate (below) or wavenumber (above the barrier)
    f_below = np.cos(al * w) * np.cosh(be * b) + (be**2 - al**2) / (2 * al * be) * np.sin(al * w) * np.sinh(be * b)
    f_above = np.cos(al * w) * np.cos(be * b) - (al**2 + be**2) / (2 * al * be) * np.sin(al * w) * np.sin(be * b)
    return np.where(below, f_below, f_above)

E = np.linspace(0.0005, 15, 150000)
E = E[np.abs(E - V0) > 1e-4]                           # avoid dividing by zero exactly at V0
allowed = np.abs(kp(E)) <= 1                           # a real k exists only where |cos(ka)| <= 1
change = np.flatnonzero(np.diff(allowed.astype(int)))
starts, ends = E[change[::2] + 1], E[change[1::2]]
print(f"Kronig-Penney crystal: period {a} nm, barriers {b} nm wide and {V0} eV high")
print("band   bottom/eV   top/eV   width/eV   gap above/eV")
for i in range(len(ends)):
    gap = f"{starts[i + 1] - ends[i]:.3f}" if i + 1 < len(starts) else "-"
    print(f"  {i + 1}      {starts[i]:6.3f}    {ends[i]:6.3f}    {ends[i] - starts[i]:6.3f}      {gap}")
print("free electrons reach the zone boundaries k = n pi/a at",
      ", ".join(f"{C * (n * np.pi / a) ** 2:.3f}" for n in (1, 2, 3)), "eV")

# The band structure E(k): invert cos(ka) = kp(E) inside the allowed energies
print("\n k a/pi   band 1   band 2   band 3   (eV)")
for ka in np.linspace(0, np.pi, 6):
    row = []
    for lo, hi in zip(starts, ends):
        grid = np.linspace(lo, hi, 20001)
        row.append(grid[np.argmin(np.abs(kp(grid) - np.cos(ka)))])
    print(f"  {ka / np.pi:4.1f}  " + "  ".join(f"{x:7.3f}" for x in row))

# Independent check: diagonalise the same crystal in a basis of plane waves exp(i(k + G)x)
m = np.arange(-60, 61)
G = 2 * np.pi * m / a
dm = m[:, None] - m[None, :]
g = 2 * np.pi * dm / a
with np.errstate(divide="ignore", invalid="ignore"):
    V = np.where(dm == 0, V0 * b / a, V0 / (1j * g * a) * (np.exp(-1j * g * w) - 1))
for ka in (0.0, np.pi):
    H = np.diag(C * (ka / a + G) ** 2) + V
    print(f"plane-wave check, k = {ka / np.pi:.0f} pi/a:", "  ".join(f"{x:.3f}" for x in np.linalg.eigvalsh(H)[:4]), "eV")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The two methods agree to the last printed digit on every band edge: 0.490, 1.541, 2.562, 6.155, 7.108 and 13.823 eV. Compare them with free electrons, which reach the zone boundaries at 1.504, 6.016 and 13.537 eV. Each gap opens around one of those energies, just as the Bragg argument predicts. The gaps shrink as the energy rises, from 1.02 to 0.95 to 0.68 eV, because a fast electron is barely disturbed by a 3 eV barrier. Away from the gaps the bands follow the free-electron parabola closely, folded back into the first Brillouin zone. The lowest band is narrow (1.05 eV) and starts at 0.49 eV rather than zero, because an electron confined mostly to one well has a zero-point energy: it is the tight-binding band of the previous section, made from the ground state of each well.

051015energy (eV)0π/awavevector kgap 1.02 eVgap 0.95 eVgap 0.68 eVband 1band 2band 3
The Kronig–Penney band structure computed above (solid curves), drawn in the first Brillouin zone, with the free-electron parabola folded into the same zone (dashed). Shaded strips are the band gaps. The crystal's bands follow the free-electron curves except near the points where two free-electron branches cross, at k = 0 and k = π/a; there the periodic potential pushes them apart and opens a gap.
ExplanationH

Metals, semiconductors and insulators

Now count electrons. Each band holds 2N2N electrons in a crystal of NN primitive cells. Fill the bands from the bottom, at absolute zero, and one of two things happens.

  • The last band is partly full. Empty states lie immediately above the highest filled ones, so an electric field can shift electrons into states with slightly more momentum in one direction: a current flows. This is a metal. Sodium, with one valence electron per primitive cell, half-fills its band. So does copper (its d bands are full; its single 4s electron half-fills an s-like band). Aluminium has three, an odd number, which can never fill whole bands.
  • The last band is exactly full, with a gap above. A full band carries no current: for every electron moving one way there is one moving the other, and there are no empty states to shift into. This is an insulator or, if the gap is small, a semiconductor. Silicon and diamond have two atoms per primitive cell and eight valence electrons per cell: exactly four bands' worth, the four bonding combinations of the sp3sp^3 orbitals, with the antibonding bands above the gap.

An even number of electrons is necessary for an insulator but not sufficient: if two bands overlap in energy, both are partly filled. Magnesium, zinc and calcium, with two valence electrons per atom, are metals for this reason. A slight overlap gives a semimetal, with few carriers of both kinds: graphite and bismuth are examples.

metalsemiconductorinsulatorsemimetalNa, Cu, AlSi, GaAsdiamond, NaClgraphite, BiE_Fsmall gaplarge gapenergy
Band filling at absolute zero. Shaded areas are filled states; the dashed line is the Fermi level. A metal has a partly filled band. A semiconductor and an insulator both have a full band (the valence band) and an empty one (the conduction band); they differ only in the size of the gap. In a semimetal two bands overlap slightly, leaving a few holes at the top of one and a few electrons at the bottom of the other.

The line between semiconductor and insulator is a matter of degree. What matters is how many electrons heat can lift across the gap. With the Fermi level near mid-gap, the occupation of the lowest conduction states is roughly e−Eg/2kBTe^{-E_g/2k_BT}, and at 300 K, where kBT=0.02585k_BT = 0.02585 eV:

MaterialGap EgE_g (eV, 300 K)Eg/2kBTE_g/2k_BTe−Eg/2kBTe^{-E_g/2k_BT}
germanium0.6612.82.9×10−62.9 \times 10^{-6}
silicon1.1221.73.9×10−103.9 \times 10^{-10}
gallium arsenide1.4227.51.2×10−121.2 \times 10^{-12}
diamond5.47105.81.1×10−461.1 \times 10^{-46}

Multiply by the roughly 101910^{19} states per cubic centimetre near a band edge and silicon has about 101010^{10} free electrons per cubic centimetre, which conducts a little; diamond has none to speak of. Exponentials turn modest differences in gap into the enormous range of resistivities we started with. The next lesson, Semiconductor materials, does this count properly and shows how impurities change it by many orders of magnitude.

ComputationC

Effective mass and density of states in three dimensions

Real crystals are three-dimensional, and their bands have shapes the free-electron model cannot match. The program takes the simplest three-dimensional tight-binding crystal, simple cubic with one orbital per atom, whose band is E(k)=−2t(cos⁡kxa+cos⁡kya+cos⁡kza)E(\mathbf k) = -2t(\cos k_xa + \cos k_ya + \cos k_za). It samples 64364^3 points of the Brillouin zone, histograms their energies into a density of states, and then fills the band to find the Fermi level for different numbers of electrons. At low filling it compares the result with a free-electron gas of the effective mass m∗=ℏ2/(2ta2)m^* = \hbar^2/(2ta^2).

python
import numpy as np

# Simple cubic crystal, one s orbital per atom, hopping t: E(k) = -2t (cos kx a + cos ky a + cos kz a)
t = 1.0                                     # eV
n = 64                                      # k-points per direction: 64^3 = 262144 states
k = 2 * np.pi * (np.arange(n) + 0.5) / n    # k a over the Brillouin zone
c = np.cos(k)
E = -2 * t * (c[:, None, None] + c[None, :, None] + c[None, None, :]).ravel()
E.sort()

# Density of states as a histogram (states per atom per eV, both spins)
counts, edges = np.histogram(E, bins=24, range=(-6, 6))
g = 2 * counts / E.size / (edges[1] - edges[0])
print("  E/eV   g(E) per atom per eV")
for gi, lo in zip(g, edges):
    print(f"{lo + 0.25:+6.2f}  {gi:5.3f} " + "#" * int(round(gi * 120)))

# Fill the band: the Fermi level is the energy below which the electrons fit (two per state)
print("\nelectrons per atom   E_F/eV   E_F - band bottom   free-electron estimate")
hbar2_over_2me = 0.0380998                  # eV nm^2
a = 0.25                                    # nm, lattice spacing
m_ratio = hbar2_over_2me / (t * a**2)       # effective mass at the band bottom, m*/m_e
for x in [0.01, 0.05, 0.2, 1.0, 1.8]:
    EF = round(float(E[int(x / 2 * E.size) - 1]), 6) + 0.0
    n_e = x / (a * 1e-9) ** 3                       # electrons per m^3
    kF = (3 * np.pi**2 * n_e) ** (1 / 3) * 1e-9     # nm^-1
    free = hbar2_over_2me * kF**2 / m_ratio
    print(f"      {x:4.2f}            {EF:+6.3f}      {EF + 6 * t:6.3f}             {free:6.3f}")
print(f"(m*/m_e = {m_ratio:.3f} for t = {t} eV and a = {a} nm)")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The density of states starts at zero at the band bottom and rises like E−Emin⁡\sqrt{E - E_{\min}}, just like free electrons, has kinks (van Hove singularities, now finite) at ±2t\pm 2t, and is symmetric about the centre. With 0.01 electrons per atom, the Fermi level sits 0.426 eV above the band bottom, close to the 0.444 eV of a free-electron gas with m∗=0.61 mem^* = 0.61\,m_e. The agreement worsens as the band fills: at half filling (one electron per atom) the free-electron estimate is wrong by more than half. And at 1.8 electrons per atom the band is nearly full and the useful description is the opposite one: a few empty states near the top.

Those empty states are holes. Near the top of a band the curvature d2E/dk2d^2E/dk^2 is negative, so electrons there have a negative effective mass and accelerate against the force on them. The collective motion of an almost-full band is exactly that of the few missing electrons treated as particles of positive charge and positive mass. Holes are not a metaphor: the Hall effect measures their positive charge directly, which in 1929 was one of the first triumphs of band theory, explaining why some metals such as zinc and beryllium show a "wrong-sign" Hall effect.

Real effective masses vary widely and can depend on direction. Electrons in gallium arsenide have m∗=0.067 mem^* = 0.067\,m_e and very high mobility, one reason it is used in fast radio-frequency transistors. Electrons in silicon sit in six equivalent valleys, each elongated, with a mass of 0.98 me0.98\,m_e along the valley and 0.19 me0.19\,m_e across it. At the other extreme, some compounds of cerium, ytterbium and uranium have f-electron bands so narrow that their carriers behave as if hundreds of times heavier than free electrons: the "heavy-fermion" metals.

ExplanationH

The professional view

Computing bands. Nobody solves Kronig–Penney for a real material. Band structures are computed with density functional theory (DFT), the subject of Density functional theory, which replaces the many-electron problem with independent electrons in an effective potential that includes, approximately, their exchange and correlation. Plane-wave codes such as Quantum ESPRESSO and VASP, and localised-orbital codes such as SIESTA and FHI-aims, all solve the same eigenvalue problem you solved above, self-consistently and in three dimensions. A Quantum ESPRESSO input for silicon looks like this (not run here; it needs the code and a pseudopotential file):

text
&CONTROL
  calculation = 'scf'
  prefix = 'si'
/
&SYSTEM
  ibrav = 2            ! face-centred cubic lattice
  celldm(1) = 10.26    ! lattice parameter in bohr (0.543 nm)
  nat = 2              ! two atoms in the primitive cell
  ntyp = 1
  ecutwfc = 30         ! plane-wave cut-off in rydberg
/
&ELECTRONS
/
ATOMIC_SPECIES
  Si  28.086  Si.pbe.UPF
ATOMIC_POSITIONS alat
  Si  0.00  0.00  0.00
  Si  0.25  0.25  0.25
K_POINTS automatic
  8 8 8 0 0 0

A second, non-self-consistent run along lines through the Brillouin zone then gives En(k)E_n(\mathbf k). Results for well over a hundred thousand compounds are already in open databases such as the Materials Project (silicon is entry mp-149), AFLOW and NOMAD, so a materials scientist usually looks a band structure up before computing one.

Tight binding is alive. Fitting a few hopping parameters to a DFT result, in the scheme of John Slater and George Koster (1954), or constructing maximally localised Wannier functions with the open-source Wannier90 code, gives small models that can describe millions of atoms, transport through devices and the effects of strain. The models used to design transistors in Semiconductor materials and in Lumenary EE's transistor scaling descend from these.

Measuring bands. Angle-resolved photoemission spectroscopy (ARPES) shines ultraviolet light or soft X-rays on a crystal and measures the energy and direction of the electrons knocked out; from those it reconstructs E(k)E(\mathbf k) of the occupied bands directly. Optical absorption measures gaps; quantum oscillations in a magnetic field measure the shape of a metal's Fermi surface. These methods return in Spectroscopy and thermal analysis.

The frontier. Since about 2005, band theory has had a new chapter: bands carry topological invariants, integers computed from how the Bloch functions unku_{n\mathbf k} twist across the Brillouin zone (which, with its opposite faces identified, is a torus). Bands with non-zero invariants must have conducting states at the surface even when the interior is insulating; these are the topological insulators of Two-dimensional and quantum materials, reviewed by Hasan and Kane in the references.

History & global context

A short history

Paul Drude's 1900 model of a metal as a gas of classical electrons explained Ohm's law and the link between electrical and thermal conductivity, but predicted a heat capacity and magnetic behaviour the experiments did not show. In 1927–28, Wolfgang Pauli and Arnold Sommerfeld replaced classical statistics with Fermi–Dirac statistics and most of those puzzles disappeared. Felix Bloch, a doctoral student of Werner Heisenberg in Leipzig, proved his theorem in 1928, and Rudolf Peierls showed how gaps and holes follow from it. Kronig and Penney's exactly solvable crystal appeared in 1931, the same year that Alan Wilson in Cambridge explained insulators and semiconductors by band filling. It took another fifteen years, and the development of pure germanium and silicon crystals, before that theory made the transistor possible in 1947.

Common misconceptionsH

Common misconceptions

Misconception

"Electrons in a metal are slowed down by bumping into the ions." In a perfectly periodic crystal a Bloch electron travels forever without scattering; the ions are already built into its wavefunction. Resistance comes from departures from perfect periodicity: lattice vibrations (phonons), impurities, vacancies, dislocations and grain boundaries. That is why very pure copper at low temperature conducts thousands of times better than at room temperature.

Misconception

"A band gap is a region of space the electron cannot enter." The gap is a range of energies that no state has. Bloch states in both bands spread through the whole crystal.

Misconception

"Insulators have no free electrons because their electrons are held tightly by atoms." Valence electrons in diamond are delocalised Bloch waves just as in a metal. Diamond insulates because its bands are exactly full, not because its electrons are stuck.

Misconception

"A hole is just a gap where an electron used to be; it does nothing on its own." A hole has a definite positive charge, effective mass and mobility, all measurable. In p-type silicon holes carry nearly all the current.

Misconception

"The effective mass is the electron's real mass changed by the crystal." The electron's mass is unchanged. The effective mass summarises how the band's curvature makes an electron respond to an applied force, with the forces from the lattice already included. It can be smaller or larger than mem_e, depend on direction, or be negative.

Exercises

Exercises

1. Electronic structure and bands

Warm-upHuman

Electron counting alone (an odd number of valence electrons per primitive cell) guarantees that a crystal has a partly filled band. For which of these solids does counting by itself prove that it is a metal? Select all that apply.

Choose all that apply

2. Electronic structure and bands

Warm-upHuman

A tight-binding ring of N=10N = 10 identical atoms, one orbital each, has energies Ej=−2tcos⁡(2πj/10)E_j = -2t\cos(2\pi j/10) for j=0,1,…,9j = 0, 1, \ldots, 9. How many distinct energy levels does it have?

3. Electronic structure and bands

CoreHuman

Silver is face-centred cubic with lattice parameter a=0.4086a = 0.4086 nm and one valence electron per atom. Using the free-electron model, compute its Fermi temperature TF=EF/kBT_F = E_F/k_B. Give the answer in thousands of kelvin, rounded to the nearest whole number.

4. Electronic structure and bands

CoreHuman

An electron sits near the top of a band, where E(k)≈Emax⁡−Bk2E(k) \approx E_{\max} - Bk^2 with B>0B > 0. A weak electric field pushes on it with force FF. Which statement is right?

Choose one

5. Electronic structure and bands

Warm-upHuman

Whose theorem says that in a periodic potential every energy eigenstate can be chosen to satisfy ψ(x+a)=eikaψ(x)\psi(x + a) = e^{ika}\psi(x)? Give the physicist's surname.

6. Electronic structure and bands

CoreComputational

Write fermi_energy(a_nm, atoms_per_cell, valence) that returns the free-electron Fermi energy in electronvolts of a metal whose cubic unit cell has lattice parameter a_nm (in nanometres), contains atoms_per_cell atoms, each contributing valence electrons.

Use n=(atoms×valence)/a3n = (\text{atoms} \times \text{valence})/a^3, kF=(3π2n)1/3k_F = (3\pi^2 n)^{1/3} and EF=ℏ2kF2/2meE_F = \hbar^2k_F^2/2m_e, with ℏ=1.054571817×10−34\hbar = 1.054571817\times10^{-34} J s, me=9.1093837015×10−31m_e = 9.1093837015\times10^{-31} kg and 1 eV =1.602176634×10−19= 1.602176634\times10^{-19} J.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

7. Electronic structure and bands

StretchComputational

Band-structure codes give E(k)E(k) at a list of points. Write effective_mass_ratio(k, E) that takes NumPy arrays of wavevectors k (in nm−1^{-1}) and energies E (in eV) sampled near a band minimum or maximum, fits a parabola E≈c2k2+c1k+c0E \approx c_2k^2 + c_1k + c_0 by least squares, and returns m∗/mem^*/m_e.

Use ℏ2/(2me)=0.0380998\hbar^2/(2m_e) = 0.0380998 eV nm2^2, so that m∗/me=0.0380998/c2m^*/m_e = 0.0380998/c_2. Near a maximum the answer must come out negative.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Further reading and resources