Build a band from localized orbitals
Keep hopping signs, Hermitian conjugation, and basis overlap explicit.
On this page
The question for today
What you’ll understand
- Derive a nearest-neighbor cosine band with a stated hopping convention.
- Use Bloch phases consistently for several orbitals.
- Recognize when an overlap matrix requires a generalized eigenproblem.
Tight binding starts from localized orbital degrees of freedom and asks how tunneling between cells spreads their atomic energies into bands. It complements the nearly-free picture by choosing a different small quantity: intersite overlap or hopping relative to relevant atomic separations. A useful tight-binding model is defined by more than a plotted cosine. It needs an orbital basis, a sign convention for hopping, and an explicit decision about whether that basis is orthonormal.
A Hamiltonian coefficient fixes the sign
Take one orthonormal orbital per site in a one-dimensional chain, with onsite energy ε₀ and a nearest-neighbor Hamiltonian matrix element −t, where t is positive. Apply the Hamiltonian to a Bloch sum whose site amplitudes are proportional to exp(ikja). Hopping from the two neighbors contributes opposite phases, whose sum is twice a cosine. This yields a band with its minimum at zero wavevector for the chosen negative off-diagonal convention.
Other authors call the actual off-diagonal element itself t, which changes the displayed sign of the cosine. Neither convention is wrong if the Hamiltonian is stated. Complex hopping requires the reverse hop to be its complex conjugate to maintain Hermiticity. Local changes of orbital phase can redistribute hopping phases, whereas phase accumulated around a closed loop can encode physical magnetic flux. An arbitrary phase assignment cannot be interpreted as a magnetic field without checking gauge-invariant loops.
Several orbitals create a matrix at each k
If each primitive cell contains multiple orbitals, Bloch sums form a small orbital-space Hamiltonian for each wavevector. Fourier-transform hopping over cell displacements, using the same sign convention as the chosen Bloch basis. One can include intracell orbital positions in the phases or leave them in the basis functions; this changes matrix representations and eigenvector phases while leaving properly computed energies invariant. Quantities involving derivatives of eigenvectors require consistent handling of that basis convention.
Atomic orbitals centered on different atoms are not automatically orthogonal. If their overlap is retained, the coefficient vector obeys a generalized eigenvalue equation with overlap matrix S. Its normalization is c†Sc=1, not c†c=1. Simply dropping S while keeping the same matrix elements changes the model. Orthogonalization can be useful, but it transforms the Hamiltonian too and may introduce longer-range matrix elements that were absent in the original nonorthogonal representation.
Truncation is a physical approximation
Longer-range hopping adds higher reciprocal harmonics and can alter extrema, curvature, or Fermi-surface topology even when its energy scale is smaller than the nearest-neighbor term. Orbitals omitted from the basis can also shift effective parameters through virtual mixing. Fitting several energies does not guarantee correct velocities or orbital weights between them. A model should therefore be validated against the observable and energy range for which it will be used, rather than judged only by whether it resembles a textbook band.
The spectral lab provides a Fourier-weight illustration. Reset and set mode count to 1 and coefficient decay to 0; predict highest-mode coefficient 1. Reset and set mode count to 3 and decay to 2; predict highest coefficient 1/9 and a smallest displayed wavelength of 2π/3. This prescribed sine sum uses sin(nx−t)/nᵖ. It is not the cosine hopping dispersion, a Bloch eigensolver, or a calculation of orbital overlap.
Work it through
A nearest-neighbor chain
Use ε₀=1.0 eV, t=0.20 eV, and a=0.4 nm. Find the energy at k=0, π/(2a), and π/a, and the bandwidth. Identify the phase between neighboring site amplitudes at the middle point.
At zero wavevector both hopping phases are one, so E=1.0−0.40=0.60 eV. The positive t and negative Hamiltonian matrix element put this state at the band minimum.
At k=π/(2a) the cosine vanishes, giving 1.00 eV. Neighboring amplitudes differ by exp(iπ/2)=i; the two neighbor contributions cancel in the energy even though hopping remains present.
At k=π/a the cosine is −1, giving 1.40 eV. The width from minimum to maximum is 0.80 eV, equal to 4t. The equivalent zone boundary at −π/a is the same reduced-zone label.
The three energies are 0.60, 1.00, and 1.40 eV, with bandwidth 0.80 eV. The middle state has a neighboring amplitude phase i.
The living notebook
Explore the model
Interactive lab · spectral
Inspect higher Fourier weights
Reset and set N=1, p=0; predict highest-mode coefficient 1. Reset and set N=3, p=2; predict 1/9 and smallest wavelength 2π/3.
- Sum at x = π/2
- 0.867
- Smallest wavelength
- 2π / 5
- Highest-mode coefficient
- 0.2
A finite sum of sin(nx − t)/nᵖ. Playback changes t; increasing the cutoff includes finer spatial scales. This is not a proof about a continuum limit or renormalization. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forThe finite sine sum illustrates truncation and spatial harmonics. Its playback is not electronic band motion, and it does not diagonalize tight-binding or overlap matrices.
Pause & reason
Check your understanding
Your turn
Try it yourself
For ε₀=0, t₁=0.20 eV and t₂=−0.04 eV, find E(0), E(π/a), and curvature at zero. Compare with t₂=0 at the same t₁.
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
The second-neighbor cosine is one at both endpoints, but its curvature contribution carries a factor of four.
Compare with the solution
E(0)=−0.40+0.08=−0.32 eV and E(π/a)=0.40+0.08=0.48 eV. Curvature at zero is 2a²(0.20−0.16)=0.08 eV·a², compared with 0.40 eV·a² without t₂. Thus the endpoint difference stays 0.80 eV while the local curvature drops by a factor of five; the full band should be inspected before inferring other extrema.
Keep the idea
Take it with you
- A tight-binding dispersion inherits its sign from the declared Hamiltonian.
- Orbital phase conventions alter representations but not consistent physical predictions.
- Overlap and longer-range hopping can matter even when endpoint energies look reasonable.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.