Lecture 2 · Part I: Quantum mechanics you need
Wavefunctions, operators and eigenvalues
What does Ĥψ = Eψ actually mean?
You will be able to explain
- what a wavefunction is, and how |ψ|² gives probabilities
- what operators, eigenfunctions and eigenvalues are
- what a measurement returns, and how an expectation value differs from an eigenvalue
- how a basis turns Ĥψ = Eψ into a matrix problem, Hc = Ec
- what NBANDS and VASP's eigenvalues are, and why they are not, in general, measured energies
- how reciprocal space labels crystal waves, what a k-point specifies, and how band edges define a gap
By Osman Goni Ridwan · Last reviewed
DFT & VASP Course · Lecture 21 / 24
Lecture 1 ended with an equation, H^ψ=Eψ, written for the electrons of one fixed geometry. Before interpreting VASP's band energies, we need to understand their labels: what reciprocal space is, what a k-point specifies, and how energies across k-points make bands and a band gap. We first define wavefunction and phase, then build those crystal pictures and use one particle in a box to explain the state ψ, the operator H^, and the energy E.
Press P to present one slide at a time, S for speaker notes.
What is a wavefunction?
A function ψ(x) whose squared magnitude, |ψ|², is a probability density.
DFT & VASP Course · Lecture 22 / 24
The state of a quantum system is described by its wavefunction, ψ. For one particle in three dimensions, ∣ψ(x,y,z)∣2dxdydz is the probability of finding the particle in the small volume dxdydz around the point (x,y,z). So ψ itself is not a probability. It can be negative, and in general it is a complex number, ψ=u+iv, whose squared magnitude ∣ψ∣2=ψ∗ψ=u2+v2 is never negative. That squared magnitude is the probability density. In one dimension, the probability of finding the particle between a and b is
P(a≤x≤b)=∫ab∣ψ(x)∣2dx.The particle must be somewhere, so the probability over all space is one: ∫∣ψ∣2dx=1. A wavefunction that satisfies this is normalized. Any solution can be normalized by dividing it by the square root of ∫∣ψ∣2dx.
A probability density is probability per unit length in this 1D model. Its value at one point is not the probability of that point. The integral sign ∫ab adds narrow strips of height ∣ψ∣2 and width dx between a and b: the area gives the probability of an interval. In 3D, add density times small volumes instead. The next slide explains the complex-number notation used for general wavefunctions.
In the figure, the second state ψ2 is negative in the right half of the box, but ∣ψ2∣2 is not: the overall sign of ψ has no meaning on its own. Multiplying ψ by −1, or by any eiθ, changes no probability. Relative signs matter only when states are added together, as a later slide shows.
What does a complex phase mean?
A complex amplitude has two components; changing its phase rotates them without changing its magnitude.
DFT & VASP Course · Lecture 23 / 24
The number i obeys i2=−1. A complex number z=u+iv has real component u and imaginary component v, drawn on perpendicular axes. Its complex conjugate is z∗=u−iv, and ∣z∣2=z∗z=u2+v2. The symbol * changes the sign of the imaginary part. It does not mean multiplication by an unspecified quantity.
Phase is the angle of this arrow. A full turn is 360° or 2π radians; a quarter turn is 90° or π/2. The complex exponential eiθ is shorthand for a unit arrow at that angle. Try 0°, 90° and 180°: the numbers are 1, i and −1, but all have squared magnitude 1.
This diagram isolates one unit amplitude to explain the arithmetic. Multiplying an entire wavefunction by the same phase factor leaves its probability density unchanged. Phase differences between components or between translated cells carry information: the next slides use them to define k. The cell-to-cell phase is a spatial relation, while a time-dependent phase describes evolution in time.
What does a k-point actually label?
k specifies how an electron wave's phase changes from one crystal cell to the next.
DFT & VASP Course · Lecture 24 / 24
A crystal repeats after a lattice translation. In this one-dimensional example the repeat distance is a. An electronic wave can extend over the whole crystal, and its value at the equivalent position in the next cell can differ by a phase factor eika. A k-point labels this pattern of phase changes. It labels an extended wave, rather than a position of an atom or an electron.
To picture a complex phase, imagine an arrow in the complex plane: its horizontal component is the real part and its vertical component the imaginary part. Multiplication by eiθ rotates the arrow through angle θ, preserving its length. Here θ=ka is the rotation per cell. The arrows in the figure compare wavefunction values at corresponding positions in different cells; they show neither particle trajectories nor motion in time. Its squared magnitude is the probability density introduced on the first slide.
Try the three presets. At Γ (Gamma), k=0, equivalent cells have the same phase. At k=41(2π/a), the phase advances by 90° per cell: the sequence is 1,i,−1,−i,1,…. At the zone edge, k=π/a, neighboring cells differ by a minus sign. These are different ways a wave can fit the same repeated structure.
The full orbital is ψnk(x)=eikxunk(x), with unk(x+a)=unk(x). The figure isolates the Bloch phase factor to make the cell-to-cell relation visible; an actual orbital also has structure inside each cell, and that structure can depend on k. Even a Γ-point orbital can vary strongly inside the cell. For any one Bloch state, ∣ψ(x+a)∣2=∣ψ(x)∣2, because the phase factor has magnitude one. Γ therefore means zero phase change per cell; it specifies neither zero energy nor an electron at rest. The full derivation of Bloch's theorem comes in Lecture 14.
What is reciprocal space?
Real space locates positions; reciprocal space locates the wavevectors that label phase patterns.
DFT & VASP Course · Lecture 25 / 24
Reciprocal space is the coordinate space of wavevectors. A plane-wave factor eikx advances in phase as x increases; for nonzero k its wavelength is 2π/∣k∣. Because kx is an angle, k has inverse-length units, conventionally Å⁻¹ with radians understood. The Bloch factor of a crystal wave uses this same k, while the periodic factor unk supplies the additional structure inside a cell.
In one dimension, adding b=2π/a to k adds a full turn to the phase change per cell: ei(k+b)a=eika. Thus k and k + b give the same cell-to-cell boundary condition. All distinct translation-phase labels can be represented within one first Brillouin zone, here [−π/a,π/a). The dark dots at integer multiples of b form the reciprocal lattice. A sampling k-point can lie anywhere within the zone; it need not coincide with one of those lattice dots.
Increase a in the figure from 2 Å to 4 Å. The real-space repeat distance doubles, while b and the width of the Brillouin zone halve. The highlighted point stays at one quarter of b. Its physical k in Å⁻¹ halves, but its phase change per cell, ka=π/2, stays at 90°.
In three dimensions there are three real-space lattice vectors ai and three reciprocal vectors bj, defined by ai⋅bj=2πδij. A k-point has three coordinates. In VASP's reciprocal-coordinate convention, (f1,f2,f3) means k=f1b1+f2b2+f3b3. Its phase change along ai is 2πfi. For example, 0.25 0 0 means a quarter turn along a1 and no phase change along a2 or a3. These are fractional reciprocal coordinates, rather than Cartesian positions in Å. Lecture 13 and Lecture 14 develop the geometry.
The dot product k⋅R multiplies corresponding vector components and adds them. The Kronecker delta δij equals 1 when i = j and 0 otherwise. Thus each reciprocal basis vector pairs with its own real-space translation to give a full turn, and with the other two to give zero phase.
How does a k-point work in three dimensions?
Three fractional k-coordinates set phase changes along three lattice translations.
DFT & VASP Course · Lecture 26 / 24
The one-dimensional picture generalizes to three lattice translations. A k-point is now a vector k=f1b1+f2b2+f3b3. Translating by a1 multiplies the wavefunction by ei2πf1, translating by a2 multiplies it by ei2πf2, and similarly for a3. Each slider controls one of these three phase changes. The two cubes rotate together so you can compare their labelled axes.
Start with a₁ only: the reciprocal coordinate is (0.25, 0, 0), and the phase changes by 90° along a1 and 0° along the other two translations. Choose All three for (0.25, 0.25, 0.25): all three translations now add 90°. The orange k-point moves away from the Γ point along the reciprocal-space body diagonal. Choose Γ to make all three phase changes zero. In every case the real-space atomic positions stay fixed.
For a general translation R=m1a1+m2a2+m3a3, with integer mi, the phase change is k⋅R=2π(f1m1+f2m2+f3m3). Thus three coordinates determine the wave's boundary condition in every cell of the infinite crystal. The orange line is the wavevector in reciprocal space, not a path taken by an electron.
This model uses a simple cubic real-space lattice. Its reciprocal vectors point along the same three directions, and its first Brillouin zone is a cube. Silicon's diamond structure has a face-centered cubic set of cell translations (an fcc Bravais lattice), with a different zone shape. For other lattices the reciprocal vectors need not be parallel to the real-space vectors; the general relation ai⋅bj=2πδij, and hence the meaning of the fractional coordinates, still holds. Lecture 13 and Lecture 14 treat these geometries.
Why does VASP need many k-points?
A mesh samples electronic states throughout the Brillouin zone; a path shows their energies along chosen directions.
DFT & VASP Course · Lecture 27 / 24
One k-point specifies one set of phase changes across the repeated cell. At that k, the electronic problem has many solutions, indexed by a band label n. To compute the density and total energy of a bulk crystal, VASP combines occupied states across the Brillouin zone. Numerically it approximates this integration using a finite set of k-points and weights. A denser mesh improves that approximation until the property of interest converges. Adding k-points samples more wavevectors; it adds no atoms to the input structure. The mesh needed depends on the cell and electronic structure.
The figure uses a square two-dimensional lattice so the mesh fits on a page. With a 4 × 4 mesh there are 16 points before symmetry reduction, each with weight 1/16. The highlighted sample (0.25,0) has a 90° phase change along the first lattice direction and zero along the second. Increasing the mesh density adds phase patterns to the calculation. A three-dimensional 4 × 4 × 4 mesh has 64 points before symmetry reduction; VASP may use fewer distinct points with larger weights when symmetry allows it. The calculation's actual points and weights appear in IBZKPT.
Switch to Band path. It follows selected lines in reciprocal space so we can plot energies against distance along that path. Here Γ = (0,0), X = (½,0), and M = (½,½) are labels for this square lattice; other lattices have different zone shapes and special points. A line path leaves most of the zone unsampled, so a band calculation normally uses a density already converged with a regular mesh. Lecture 15 explains integration and convergence, and Lecture 29 develops band-structure calculations.
What are a band and a band gap?
A band follows an energy across k; a gap separates the filled valence bands from the empty conduction bands.
DFT & VASP Course · Lecture 28 / 24
At one k-point there is a set of allowed orbital energies ε1k,ε2k,…. Follow one energy branch as k changes and you trace an energy band, εn(k). In a three-dimensional crystal it is a function over a three-dimensional zone; a familiar band plot shows only a path through it. The horizontal axis labels wave patterns in reciprocal space. The vertical axis is energy.
For a semiconductor or insulator at zero temperature, the highest filled bands are the valence bands; the lowest empty ones are the conduction bands. The band gap of the band description is the difference between the lowest conduction energy anywhere in the zone (the CBM) and the highest valence energy anywhere in the zone (the VBM):
Eg=kminεc(k)−kmaxεv(k).The Fermi level EF is the chemical potential governing occupations. In a zero-temperature band model, states below it are occupied and states above it empty; in a metal it cuts through a partially filled band. In an insulator it can lie inside the gap. This introduces occupancy, rather than assigning a different physical position to each band or k-point.
In the Direct gap model, these extrema occur at the same k. In the Indirect gap model, they occur at different k-points. Notice that the separation of the two energies at the selected k can be larger than the global gap. A single k-point, or even a high-symmetry path through a real three-dimensional zone, can miss the extrema. In the Metal model, the Fermi level cuts a partially filled band, so there is no insulating gap at that filling.
The curves here are teaching models. Applied to VASP's Kohn–Sham eigenvalues, the formula gives a Kohn–Sham band gap. A measured fundamental gap concerns electron addition and removal, and an optical gap also involves electron–hole excitations. Those quantities can differ from the Kohn–Sham gap, as the later slide in this lecture cautions and Lecture 30 explains in detail. For now, identify the axes, the band index, and the two edges defining a band gap.
What is an operator?
A rule that turns one function into another. Every observable has one.
| Observable | Operator | What it does to ψ(x) |
|---|---|---|
| Position | x^ | Multiplies by x: x^ψ=xψ(x) |
| Momentum | p^=−iℏdxd | Takes the slope, times −iℏ |
| Kinetic energy | T^=−2mℏ2dx2d2 | Takes the curvature, times −ℏ2/2m |
| Total energy | H^=T^+V(x) | Kinetic plus potential energy: the Hamiltonian |
DFT & VASP Course · Lecture 29 / 24
An operator, written with a hat, takes a function and returns another function. Position simply multiplies by x; the potential energy V(x), like position, just multiplies. Momentum differentiates. Why? A particle of momentum p has, by de Broglie's relation, a wave with wavelength h/p, that is eikx with p=ℏk. The operator that pulls the number ℏk out of eikx is −iℏd/dx, since −iℏdxdeikx=ℏkeikx. Kinetic energy is p2/2m, so its operator differentiates twice: T^=p^2/2m=−2mℏ2dx2d2. In three dimensions d2/dx2 becomes ∇2=∂2/∂x2+∂2/∂y2+∂2/∂z2.
For a single plane wave, ℏk is its momentum. A crystal Bloch orbital generally contains many plane-wave components. Its k labels translation phases (crystal momentum), rather than a definite mechanical momentum of that full orbital.
Every observable quantity has an operator, and these operators share two properties. They are linear: A^(aψ1+bψ2)=aA^ψ1+bA^ψ2, which is what will let us turn quantum mechanics into linear algebra at the end of this lecture. And they are Hermitian, which a later slide explains.
What is an eigenvalue equation?
An eigenfunction comes back from the operator unchanged, apart from a factor: the eigenvalue.
- Operator: for example p^ or H^.
- Eigenfunction (eigenstate) of that operator.
- Eigenvalue: a number, the same everywhere in space.
- An eigenfunction of momentump^eikx=−iℏ(ik)eikx=ℏkeikx
- Not an eigenfunctionp^sinkx=−iℏkcoskx: a different function comes back.
DFT & VASP Course · Lecture 210 / 24
Most functions are changed by an operator into a different function. A few special ones come back as themselves, multiplied by a constant. They are the eigenfunctions of that operator, and the constants are its eigenvalues. "Eigen" is German for "own": the function's own value.
The plane wave eikx is an eigenfunction of momentum with eigenvalue ℏk: a state of definite momentum. The function sinkx is not, because differentiating it gives a cosine. It is, however, a sum of two plane waves, sinkx=(eikx−e−ikx)/2i, with momenta +ℏk and −ℏk; a later slide shows what measuring the momentum of such a sum gives. Notice that sinkx is an eigenfunction of the kinetic-energy operator, with eigenvalue ℏ2k2/2m: both of its plane waves have the same kinetic energy. Keep the plane wave in mind; in Lecture 17 VASP's whole basis is made of them.
What does Ĥψ = Eψ ask for?
The states with a definite energy: Ĥ returns them multiplied by a number, E.
- Kinetic energy: large where ψ curves sharply.
- Potential energy, point by point.
- Energy eigenvalue of state n.
- Many solutionsψ1,ψ2,ψ3,… with energies E1≤E2≤E3≤…
- Ground stateThe lowest one. Higher ones are excited states.
DFT & VASP Course · Lecture 211 / 24
The time-independent Schrödinger equation asks for states with definite energy and for their energies. Under the time-dependent equation, an energy eigenstate acquires a phase, ψn(x)e−iEnt/ℏ, so its probability density stays constant in time. That is what stationary means. A stationary state can still have kinetic energy or carry a probability current; a time-independent density does not imply a classical particle at rest.
There are usually many solutions, ordered by energy. The lowest is the ground state, the state DFT is built to describe; the others are excited states. The equation also shows where kinetic energy comes from: the second derivative is large where ψ bends sharply, so a wavefunction squeezed into a small region, or one with many wiggles, has high kinetic energy. On the next slide, boundary conditions decide which solutions exist.
Why does energy come in steps?
Only standing waves that vanish at both walls are allowed, so only certain energies are.
DFT & VASP Course · Lecture 212 / 24
Put a particle between two infinitely high walls a distance L apart. Inside, V=0, and the Schrödinger equation reads −2mℏ2ψ′′=Eψ. Its general solution is ψ=Asinkx+Bcoskx with E=ℏ2k2/2m. The particle cannot be inside an infinite wall, so ψ must vanish at both walls:
- ψ(0)=0 removes the cosine: B=0.
- ψ(L)=AsinkL=0 requires kL=nπ: a whole number of half-wavelengths must fit in the box.
- Normalizing, ∫0LA2sin2(nπx/L)dx=A2L/2=1, gives A=2/L.
So
ψn(x)=L2sinLnπx,En=2mℏ2(Lnπ)2=8mL2n2h2,n=1,2,3,…The energy is quantized not because anything forbids other energies in principle, but because only these waves fit. State n has n−1 nodes and energy n2E1. Two lessons carry over to real materials. More nodes mean more curvature, so more kinetic energy. And confinement costs energy: halve the box and every level rises four times. For an electron in a box 1 Å long, about the size of an atom, E1=h2/8meL2=37.6 eV, the same order as the energies that bind electrons in atoms. Use the slider to see how fast the levels rise as the box shrinks.
What if the state is not an eigenstate?
Each measurement still gives one eigenvalue; the coefficients decide how often.
DFT & VASP Course · Lecture 213 / 24
A valid initial state can be expanded as a superposition, ψ=∑ncnψn. The time-dependent Schrödinger equation evolves each term with its own phase. If nonzero terms have different energies, the sum has no definite energy. Different states sharing the same eigenvalue are called degenerate; a superposition within that degenerate energy space still has definite energy. The three box levels here have distinct energies. Measuring their superposition gives En with probability ∣cn∣2, and these probabilities sum to one for a normalized state.
Try it: set c1 and c2 and press Measure once a few times. Every result is E1 or 4E1, never the average. Press Measure 100 times and the bars approach ∣c1∣2 and ∣c2∣2.
Notice also that the probability density of the superposition is not just the sum of the two densities. For real coefficients,
∣c1ψ1+c2ψ2∣2=c12ψ12+c22ψ22+2c1c2ψ1ψ2.The cross term is interference. In the box, ψ1ψ2 is positive in the left half and negative in the right half, so with c1c2>0 the particle is more likely on the left, and flipping the sign of c2 moves it to the right. That is why the relative sign, or phase, of the parts of ψ matters.
Is an expectation value a possible result?
Not necessarily. In an eigenstate every measurement gives it; otherwise it is an average.
- Expectation value of A^ in a normalized state ψ.
- The same number: each eigenvalue weighted by its probability.
- Eigenstate ψ₂Every measurement gives 4E1, and ⟨E⟩=4E1.
- Equal superposition of ψ₁ and ψ₂Measurements give E1 or 4E1; ⟨E⟩=2.5E1, which never occurs.
DFT & VASP Course · Lecture 214 / 24
The average of many measurements on identically prepared states is the expectation value. The two forms on the slide are the same number. Insert ψ=∑ncnψn into the integral and use A^ψn=anψn:
∫ψ∗A^ψdx=m,n∑cm∗cnan∫ψm∗ψndx=n∑∣cn∣2an,because the eigenstates are orthonormal (∫ψm∗ψndx is 1 for m=n and 0 otherwise; a later slide explains why). The integral form is the one you compute; the sum form says what it means.
If ψ is an eigenstate of A^, only one term survives and every measurement gives the eigenvalue. Otherwise the average can fall between the possible results. For the equal superposition, ⟨E⟩=21E1+21⋅4E1=2.5E1, a value no single measurement ever gives. This distinction returns in Lecture 3: the energy of a trial wavefunction is an expectation value, ⟨ψ∣H^∣ψ⟩, and the sum form shows at once that it can never fall below the lowest eigenvalue, the ground-state energy.
Why must observables be Hermitian?
Hermitian operators have real eigenvalues and orthogonal eigenstates that form a complete set.
- Real eigenvaluesEvery measured value is a real number.
- Orthogonal eigenstates∫ψm∗ψndx=δmn once normalized: 1 if m=n, otherwise 0.
- Expansionψ=∑ncnψn with cn=∫ψn∗ψdx.
DFT & VASP Course · Lecture 215 / 24
An operator is Hermitian if it can be moved from one function to the other inside an integral, with a complex conjugate:
∫f∗(A^g)dx=∫(A^f)∗gdxfor all f,g.Multiplication by a real function, such as x or V(x), obviously has this property. For p^=−iℏd/dx it follows from integrating by parts, with the boundary terms vanishing because the wavefunctions go to zero far away; the factor −i is exactly what makes it work. Three consequences make the whole machinery of quantum mechanics work:
- Real eigenvalues. Every possible measurement result is a real number.
- Orthogonal eigenstates. Eigenstates with different eigenvalues are orthogonal, and degenerate ones can always be chosen orthogonal. The box states are an example: ∫0Lψ1ψ2dx=0.
- Completeness. The eigenstates form a complete set: any reasonable state can be expanded as ψ=∑ncnψn. Orthonormality then makes each coefficient a simple overlap integral. Multiply the expansion by ψm∗ and integrate: every term but one vanishes, leaving cm=∫ψm∗ψdx.
Proof of the first two properties
Let A^ψn=anψn. Take f=g=ψn in the definition: the left side is an∫∣ψn∣2dx and the right side is an∗∫∣ψn∣2dx. So an=an∗: the eigenvalue is real.
Now take f=ψm and g=ψn. The left side is an∫ψm∗ψndx and the right side, using that am is real, is am∫ψm∗ψndx. Subtracting, (an−am)∫ψm∗ψndx=0. If an=am, the integral must vanish: the two eigenstates are orthogonal.
What does ⟨ψ|Ĥ|ψ⟩ mean?
Bra–ket notation is shorthand for the integrals on the previous slides.
| Bra–ket | Integral | Meaning |
|---|---|---|
| ⟨ψ∣ψ⟩=1 | ∫ψ∗ψdx=1 | Normalization |
| ⟨ψm∣ψn⟩=δmn | ∫ψm∗ψndx | Orthonormality |
| cn=⟨ψn∣ψ⟩ | ∫ψn∗ψdx | How much of ψn is in ψ |
| ⟨ψ∣H^∣ψ⟩ | ∫ψ∗H^ψdx | Energy expectation value |
| ⟨ϕi∣H^∣ϕj⟩ | ∫ϕi∗H^ϕjdx | A matrix element, Hij |
DFT & VASP Course · Lecture 216 / 24
Electronic-structure papers and the VASP documentation write states as kets, ∣ψ⟩, and their complex conjugates as bras, ⟨ψ∣. A bra next to a ket means: multiply the conjugate of the first function by the second and integrate over all coordinates (and sum over spin, if there is spin). An operator in the middle acts on the ket first. Nothing new is introduced; the notation only saves writing, and it makes the analogy with vectors visible: ⟨ψm∣ψn⟩ behaves like a dot product, and orthonormal states behave like perpendicular unit vectors. The last row is the bridge to the next slide.
How does a computer solve Ĥψ = Eψ?
In a finite basis, ψ becomes a list of coefficients and Ĥ a matrix: Hc = Ec.
- Basis expansion: M known functions ϕj, here orthonormal.
- Hamiltonian matrix, Hij=⟨ϕi∣H^∣ϕj⟩: numbers a computer can store.
- Eigenvector: the coefficients of one state.
- Eigenvalue: its energy. An M×M Hermitian matrix has M eigenvalues, counting degeneracy.
DFT & VASP Course · Lecture 217 / 24
A computer cannot store a function at infinitely many points. It can store numbers. So expand the unknown state in a set of known basis functions, ψ=∑jcjϕj, and let the coefficients carry the information. Insert the expansion into H^ψ=Eψ and take the overlap with one basis function, ϕi:
j∑⟨ϕi∣H^∣ϕj⟩cj=Ej∑⟨ϕi∣ϕj⟩cj=Eci.The last step uses orthonormality. Doing this for every i gives M equations, which together are the matrix equation on the slide, Hc=Ec. Matrix multiplication means that row i gives ∑jHijcj=Eci. A matrix is a rectangular array of numbers; an eigenvector is the list of coefficients that this multiplication returns multiplied by E. The differential equation has become a problem of linear algebra. Because H^ is Hermitian, the matrix is Hermitian too (Hji=Hij∗), and standard library routines diagonalize it, returning all M eigenvalues and eigenvectors at once.
With a complete, infinite basis this would be exact. In practice the basis is finite, so the answer is approximate. How good it is, and what happens when the basis functions are not orthogonal (the right side becomes E∑jSijcj, with the overlap matrix Sij=⟨ϕi∣ϕj⟩, so Hc=ESc), is the subject of Lecture 3. VASP's basis is plane waves, and ENCUT sets how many (Lecture 17). Even so, VASP's PAW method introduces an overlap operator S^, so what it actually solves has the form (H^−εS^)ψ=0; Lecture 18 explains why.
What does VASP mean by a band?
Each k-point gives a set of orbital energies; following an energy branch across k traces a band.
- An operator: the Kohn–Sham Hamiltonian, rebuilt at every SCF step (Lecture 9).
- An eigenfunction: the orbital of band n at k-point k.
- An eigenvalue: a number, its energy. EIGENVAL lists them all.
- This lectureWhat an operator, an eigenfunction and an eigenvalue are, using one particle in a box.
- LaterThe full Bloch theorem (Lecture 14), k-point convergence (Lecture 15), and interpreting measured gaps (Lecture 30).
DFT & VASP Course · Lecture 218 / 24
After every run, VASP writes the file EIGENVAL, which lists the Kohn–Sham eigenvalues for every band at every k-point. Each of those numbers solves an equation of the form on the slide: an operator acts on a function and returns the same function multiplied by a number. Lecture 0 called this solving the Kohn–Sham equations. This lecture explains the vocabulary, with a system small enough to solve by hand.
Where do these objects live in VASP?
Orbitals in WAVECAR, eigenvalues in EIGENVAL and OUTCAR, their number set by NBANDS.
| In this lecture | In VASP | Where to find it |
|---|---|---|
| Operator H^ | Kohn–Sham Hamiltonian H^KS, rebuilt every SCF step | Lecture 9, Lecture 12 |
| Eigenfunction ψn | Orbital ψnk: band n, k-point k | WAVECAR |
| Eigenvalue En | Band energy εnk, in eV | EIGENVAL, OUTCAR |
| Number of states solved for | NBANDS: occupied bands plus some empty ones | INCAR, OUTCAR |
| Probability density ∣ψ∣2 | Density n(r)=∑nkwkfnk∣ψnk∣2, with k-point weights wk | CHGCAR |
DFT & VASP Course · Lecture 219 / 24
NBANDS sets how many Kohn–Sham orbitals VASP computes at each k-point. It needs at least all the occupied ones plus one empty band, and its default adds a margin, which is a safe choice in most cases. Empty bands are needed because the iterative solver converges better with them, and because properties such as the band gap or optical spectra need the lowest unoccupied states. The number of electrons, NELECT, is taken from the POTCARs for a neutral cell: the valence electrons per atom, ZVAL, times the number of atoms of each species.
One orbital is not the electron density. The density adds up the occupied orbitals, each weighted by its occupation fnk (up to two electrons per orbital without spin polarization) and by the weight wk of its k-point; the weights add up to one. In OUTCAR, every k-point gets a block that starts with band No. and lists band energies and occupations.
Are VASP's eigenvalues measurable energies?
Not in general: they belong to an auxiliary problem, not to the real interacting electrons.
- HOMO: highest occupied molecular orbital. For a finite system with the exact functional, its eigenvalue equals minus the first ionization energy I.
- What they areEigenvalues of the Kohn–Sham Hamiltonian, a stand-in system of non-interacting electrons (Lecture 9).
- Still usefulOften a fair, qualitative guide to single-particle energies.
- Watch outThe band gap they give is not the measured gap (Lecture 30).
DFT & VASP Course · Lecture 220 / 24
Everything in this lecture so far was exact quantum mechanics for one particle: each eigenvalue is a possible result of an energy measurement. The Kohn–Sham equations that VASP solves look the same, but their Hamiltonian describes a fictitious system of non-interacting electrons, constructed to have the same density as the real, interacting ones. Its eigenvalues are energies of that fictitious system.
For the exact functional, a useful exact result for a finite system is that its highest occupied eigenvalue equals minus its first ionization energy. In a periodic cell even this link cannot be used directly, because the eigenvalues have no vacuum level to be measured from. Other levels have no general exact identification with measured addition or removal energies, and in particular the Kohn–Sham gap differs from the true gap even with the exact functional, by a term called the derivative discontinuity (Lecture 30).
In practice, Kohn–Sham eigenvalues often approximate the energies of adding or removing an electron well enough to be qualitatively useful: band dispersions, orderings and densities of states usually look right, while band gaps come out too small. So read band structures with care until Lecture 29 and Lecture 30.
Ionization energy is the energy required to remove an electron to vacuum, I=E(N−1)−E(N), with the remaining system allowed to relax electronically. LUMO means lowest unoccupied molecular orbital. The HOMO–LUMO difference is an orbital-energy gap for a finite system; it is generally distinct from the interacting addition/removal gap. The term derivative discontinuity names an extra contribution when the electron number crosses an integer; its derivation is deferred to Lecture 30.
Lab 2: where are the eigenvalues in your Lab 0 output?
No new run: predict NBANDS and the occupied bands, then check them in Lab 0's OUTCAR and EIGENVAL.
Predict first
- NELECT: 2 silicon atoms × 4 valence electrons = 8.
- Occupied bands: 8 electrons, 2 per band, so 4 at every k-point.
- Default NBANDS: max(21(8+2)+max(1,3), ⌊0.6×8⌋)=8, raised if needed to a multiple of the cores you ran on.
Then check
grep NBANDS OUTCARandgrep NELECT OUTCAR.grep -A 8 "band No." OUTCAR(8 = your NBANDS): band energies and occupations per k-point.- Count the bands with occupation 2 and those with 0. Find the same k-point in EIGENVAL.
DFT & VASP Course · Lecture 221 / 24
VASP's default for a calculation without spin polarization is
NBANDS=max(21nint(Nelect+2)+max(21Nions, 3), int(53Nelect)),where nint rounds to the nearest integer and int rounds down. For silicon, Nelect=8 and Nions=2: the first argument is 5 + 3 = 8 and the second is 4, so NBANDS = 8. On several cores, VASP increases NBANDS to suit the parallel setup; with the defaults NCORE = 1 and KPAR = 1, until it is divisible by the number of cores. So if you ran Lab 0 on 12 cores, expect a different number, and work out why.
With ISMEAR = 0 and a small SIGMA, the four lowest bands should be nearly fully occupied and the rest nearly empty, because silicon has a gap. Find the reciprocal coordinates of one k-point and its list of band energies. The gap across all sampled k-points is the lowest empty-band energy minus the highest occupied-band energy; these two energies can come from different k-points. A coarse Lab 0 mesh may miss silicon's band edges, so treat this as an exercise in reading the output, rather than a converged gap calculation. Lecture 15 and Lecture 29 develop the sampling. EIGENVAL is in the same Lab 0 run directory.
Which misconceptions should you drop now?
Six confusions about crystal waves and eigenvalues. Click a card to turn it over.
“ψ is the probability of finding the particle.”
|ψ|² is a probability density. ψ itself can be negative or complex.
“The expectation value is the most likely result.”
It is an average. In a superposition it may lie between eigenvalues and never be measured.
“The sign of ψ means nothing.”
Alone it does not, but relative signs between added states change |ψ|² through interference.
“One orbital is the electron density.”
The density adds up all occupied orbitals, each weighted by its occupation.
“Every VASP eigenvalue is a measured energy.”
Kohn–Sham eigenvalues belong to an auxiliary system and need careful interpretation.
“A k-point is a position in the crystal.”
It labels a wave's phase change between repeated cells. At each k-point there are many orbitals, and each extends through the crystal.
DFT & VASP Course · Lecture 222 / 24
Check yourself
One question per outcome. Answer aloud, then click to compare.
What is the difference between ψ and |ψ|²?
ψ is the amplitude, possibly negative or complex. |ψ|² is the probability density; integrated over a region, it gives a probability.
In A^ψ=aψ, name each symbol.
A^ is an operator, ψ an eigenfunction of it, and a the eigenvalue: a number.
ψ = (ψ₁ + ψ₂)/√2 in a box. What can an energy measurement give, and what is ⟨E⟩?
E₁ or 4E₁, each with probability ½. ⟨E⟩ = 2.5 E₁, which never occurs.
How does a basis turn Ĥψ = Eψ into something a computer can solve?
ψ = Σ c_j φ_j turns it into Hc = Ec in an orthonormal basis (Hc = ESc otherwise), with H_ij = ⟨φ_i|Ĥ|φ_j⟩.
What does NBANDS set, and are its eigenvalues measured energies?
The number of Kohn–Sham orbitals per k-point. Their eigenvalues belong to an auxiliary problem, so not in general.
What does k label, and how do bands define a gap?
k labels cell-to-cell phase changes. A band follows an energy across k. The global gap is the conduction minimum minus the valence maximum; those edges can occur at different k-points.
DFT & VASP Course · Lecture 223 / 24
What should you remember?
State → wavefunction; observable → operator; possible result → eigenvalue.
- |ψ|² is a probability density, and a normalized ψ integrates to one.
- An eigenfunction comes back from its operator times a number; Ĥψ = Eψ asks for the states of definite energy.
- A measurement gives an eigenvalue, with probability |c_n|²; the expectation value is the average and may never be measured.
- In an orthonormal basis, Ĥψ = Eψ becomes Hc = Ec (in general Hc = ESc). VASP solves such a problem for NBANDS orbitals at every k-point.
- Kohn–Sham eigenvalues are not, in general, measurable excitation energies.
- k labels cell-to-cell phase changes in reciprocal space. A mesh samples the zone; a band follows an energy across k, and its global gap uses the conduction minimum and valence maximum.
Next: Lecture 3 asks how to approximate a state you cannot solve exactly, and how good the answer is.
DFT & VASP Course · Lecture 224 / 24
Key terms
- Wavefunction
- The function ψ that describes the state of a quantum system. It is in general complex; its squared magnitude ∣ψ∣2 is a probability density.
- Probability density
- ∣ψ(r)∣2=ψ∗ψ: the probability per unit volume of finding the particle at r. Integrated over a region, it gives the probability of finding the particle there.
- Normalization
- The condition ∫∣ψ∣2dτ=1: the particle is certainly somewhere.
- Complex phase
- The angle of a complex amplitude. Multiplication by eiθ rotates the amplitude while preserving its squared magnitude. Relative phases affect interference; a common overall phase leaves probabilities unchanged.
- k-points
- Wavevectors labelling the phase change of an electronic wave under a crystal translation: ψnk(r+R)=eik⋅Rψnk(r). Each k-point has many band states; a weighted mesh samples these states across the Brillouin zone.
- Reciprocal space
- The coordinate space of wavevectors, with inverse-length units. Reciprocal basis vectors obey ai⋅bj=2πδij; a fractional reciprocal coordinate fi gives a phase change of 2πfi along the real-space lattice vector ai.
- First Brillouin zone
- The Wigner–Seitz primitive cell of the reciprocal lattice, centered at Γ (k = 0). It represents the distinct crystal translation-phase labels, with equivalent boundary points identified. For a 1D lattice of period a it spans −π/a to +π/a.
- Energy band
- An orbital-energy branch εn(k) followed across reciprocal space. At a fixed k-point there are many energies, indexed by n; a band plot traces these branches along a chosen path through the Brillouin zone.
- Band gap
- For an insulating band filling, the conduction-band minimum minus the valence-band maximum across the entire Brillouin zone. The gap is direct if both extrema occur at the same k-point, and indirect if they occur at different k-points. A Kohn–Sham gap can differ from measured fundamental and optical gaps.
- Fermi level
- The chemical potential governing occupations. In a zero-temperature band model, states below it are occupied and states above it are empty; a metal has a partially filled band crossing this level.
- Hamiltonian
- The operator for the total energy of a system. For a material it collects the kinetic energies of the electrons and nuclei and all their Coulomb interactions.
- Eigenvalue equation
- A^ψ=aψ: the operator returns the same function multiplied by a number. ψ is an eigenfunction (eigenstate) and a its eigenvalue. The Schrödinger equation H^ψ=Eψ is the eigenvalue equation of the energy.
- Ground state
- The lowest-energy state of a system. Ordinary DFT is a theory of the ground state; excited states need extensions of the theory or other methods.
- Superposition
- A sum of states, ψ=∑ncnψn. Measuring an observable whose eigenstates are the ψn gives eigenvalue an with probability ∣cn∣2.
- Expectation value
- ⟨A⟩=⟨ψ∣A^∣ψ⟩=∫ψ∗A^ψdτ for a normalized state: the average of many measurements, equal to ∑n∣cn∣2an. Only in an eigenstate is it guaranteed to be an eigenvalue, the result of every measurement.
- Hermitian operator
- An operator whose expectation values are always real. Its eigenvalues are real, its eigenstates can be chosen orthonormal, and they form a complete set, so any state can be expanded in them. Observables are Hermitian.
- Orthonormal
- A set of functions with ⟨ψm∣ψn⟩=δmn: each normalized, and any two different ones orthogonal (overlap zero).
- Basis set
- A set of known functions ϕj in which an unknown state is expanded, ψ=∑jcjϕj. It turns H^ψ=Eψ into a matrix problem for the coefficients; a finite basis gives an approximate answer. VASP's basis is plane waves.
- Operator
- A rule that turns one function into another, written with a hat, A^. In quantum mechanics every observable has a linear, Hermitian operator: position multiplies by x, momentum is −iℏd/dx, energy is the Hamiltonian.
VASP tags in this lecture
ENCUT- Cutoff energy of the plane-wave basis set, in eV. The VASP Wiki strongly recommends setting it explicitly in every INCAR, and checking the convergence of the quantity you report.Default: the largest ENMAX among the POTCAR files · VASP Wiki: ENCUT
NBANDS- Total number of Kohn–Sham orbitals (bands) per k-point. VASP needs at least all occupied bands plus one empty one, and on several cores it raises NBANDS to a convenient multiple.Default: max(nint(NELECT + 2)/2 + max(NIONS/2, 3), int(0.6 NELECT)) · VASP Wiki: NBANDS
NELECT- Number of electrons. By default it is set from the POTCARs (ZVAL of each species times its number of atoms) for a neutral cell; set it only for a charged calculation.Default: the valence electrons of the POTCARs (neutral cell) · VASP Wiki: NELECT
Further reading
- R. W. Field, 5.61 Physical Chemistry, Lecture 5: “Begin quantum mechanics: free particle and particle in a 1D box”, MIT OpenCourseWare (Fall 2017), CC BY-NC-SA. · MIT OCW
- R. W. Field, 5.61 Physical Chemistry, Lecture 12: “Looking backward before first hour exam: postulate”, MIT OpenCourseWare (Fall 2017), CC BY-NC-SA. · MIT OCW
- VASP Wiki, “Crystal momentum” (Bloch translation phases, reciprocal vectors and Brillouin-zone sampling). · vasp.at/wiki
- VASP Wiki, “KPOINTS” (every format, the mesh formulas, and which mesh suits which lattice). · vasp.at/wiki
- VASP Wiki, “BANDGAP” (valence and conduction band edges, fundamental and direct band gaps). · vasp.at/wiki