Relativistic Quantum Mechanics for Readers Who Know Only Nonrelativistic Quantum Mechanics — From Schrödinger to Klein–Gordon and Dirac, and Relativistic Versions of the Free-Electron Gas and the Thomas–Fermi Model

— What do Lorentz symmetry, gauge symmetry, the Klein–Gordon equation, and the Dirac equation actually mean? Explanation and derivations for the companion simulator (Relativistic-QM-simulator-en.html)

Lecture 7 of Computational Materials Science (slides 29–31) introduced the facts that the Schrödinger equation, being first order in time and second order in space, does not satisfy the requirement of relativity (Lorentz symmetry), that turning E² = p²c² + m²c⁴ into operators gives the Klein–Gordon equation, and that Dirac rewrote it to first order and arrived at a four-component spinor, spin, and antiparticles. This explanation is the sequel. To the question “a relativistic equation should also give eigenvalues and eigenfunctions just as the Schrödinger equation does, but what exactly changes?”, we answer with the free electron, the infinite well, the free-electron gas, and the Thomas–Fermi model, the examples most familiar from nonrelativistic physics.

To state the conclusion first, what changes is (i) the dispersion relation E(p) (parabola → hyperbola, with a negative-energy branch), (ii) the number of components of the wave function (1 → 4: spin and antiparticles), (iii) the way boundary conditions are imposed (the Dirac equation is first order, so “ψ = 0 at the wall” cannot be used), and (iv) the coupling to the electromagnetic field (g = 2 and the spin–orbit interaction come out automatically from gauge symmetry). In the free-electron gas, the way of counting states in k space does not change; merely replacing E(k) changes the density dependence of the density of states, the Fermi energy, the pressure, and the exchange energy. For conduction electrons in a metal (xF = ħkF/mc ≈ 0.005) the correction is 10−5 or less; at the center of a white dwarf (xF ≳ 1) the density dependence of the pressure changes and the Chandrasekhar limiting mass appears; and in the inner shells of heavy atoms (kF ≈ 2.66Z a.u.) the relativistic factor of the exchange energy enters Dirac–Kohn–Sham calculations.

Roadmap of this explanation §1 Lorentz symmetry: why the Schrödinger equation is not enough. §2 The Klein–Gordon equation: negative-energy solutions and the true nature of the “negative probability” problem. §3 The Dirac equation: factorization demands 4×4 matrices, and spin and antiparticles emerge. §4 Gauge symmetry: the freedom of phase demands an electromagnetic field, and Dirac + minimal coupling gives g = 2, the spin–orbit interaction, and the Darwin term. §5 A particle in a box: KG and Dirac levels, the MIT boundary condition, and Klein’s paradox. §6 The relativistic free-electron gas: density of states, Fermi energy, pressure, the Chandrasekhar limit, and the TF kinetic-energy functional. §7 Relativistic exchange energy: the relativistic version of Dirac exchange, and the sign reversal. §8 The relativistic Thomas–Fermi atom: the Vallarta–Rosen equation and the divergence at a point nucleus. §9 Correspondence table. §10 Implementation notes. §11 References.
  1. Lorentz Symmetry
  2. The Klein–Gordon Equation
  3. The Dirac Equation
  4. Gauge Symmetry and Minimal Coupling
  5. A Particle in a Box
  6. The Relativistic Free-Electron Gas
  7. Relativistic Exchange Energy
  8. The Relativistic Thomas–Fermi Model
  9. Correspondence Table
  10. Implementation Notes
  11. References

1. Lorentz Symmetry — Why Is the Schrödinger Equation Not Enough?

1.1 The starting point of relativity is an “invariant”

Special relativity starts from the statement that “the speed of light c is the same in every inertial frame,” and as a consequence says that the quantities that do not change when we switch inertial frames (a Lorentz transformation) are

(1) s2=c2t2−x2−y2−z2, E2−p2c2=m2c4

The left one is the “distance” in spacetime, and the right one is the “length” of the four-vector (E/c, p) formed by energy and momentum; in both, the time component and the space components enter on an equal footing (differing only in sign). This is what was meant in slide 29 of Lecture 7 by “time and space are not completely independent but correlated,” and an equation that treats time and space separately conflicts with the requirement of relativity.

1.2 The Schrödinger equation is Galilean invariant, not Lorentz invariant

The way quantum mechanics is built is to “substitute operators into a relation between E and p”:

(2) E→iℏ∂∂t, p→−iℏ∇, E=p22m+V ⟹ iℏ∂ψ∂t= (−ℏ22m∇2+V)ψ

Since the starting point is the nonrelativistic E = p²/2m, the resulting equation is first order in time and second order in space. Under the coordinate change x′ = x − vt (a Galilean transformation) the equation can be returned to the same form by adjusting a phase factor, but under a Lorentz transformation time and space mix, so the form cannot be preserved. “An equation has a symmetry” means that it keeps the same form when rewritten in the transformed coordinates (and therefore gives the same physics in all inertial frames). In this sense the Schrödinger equation does not have Lorentz symmetry. For practical purposes v/c ≪ 1 is enough, but where effects of order (v/c)² become visible (0.58² for the 1s electron of gold), a relativistic equation is needed.

1.3 What is the goal of “relativistic quantum mechanics”?

If the relation on the right of Eq. (1) is turned into operators, we get an equation in which time and space enter with the same order of derivative. There are two ways to do this: using the squared relation as it is gives the Klein–Gordon equation (§2), and taking the square root to reduce it to first order gives the Dirac equation (§3). In both cases the free-particle solution is the plane wave ei(p·r − Et)/ħ, and the eigenvalues are E = ±√(p²c² + m²c⁴). In that they “give eigenvalues and eigenfunctions,” they are the same as the Schrödinger equation; the differences are how the eigenvalues are arranged (the dispersion relation) and the number of components of the eigenfunction. ▶ Simulator ①: Dispersion relation

2. The Klein–Gordon Equation — Negative-Energy Solutions and the True Nature of “Negative Probability”

2.1 Derivation and the plane-wave solution

(3) E2=p2c2+m2c4 ⟹ −ℏ2∂2ψ∂t2 =−ℏ2c2∇2ψ+m2c4ψ or (1c2∂2∂t2−∇2+m2c2ℏ2)ψ=0

It is second order in time and second order in space, and is Lorentz symmetric (lecture slide 30). Inserting a plane wave gives E² = p²c² + m²c⁴, that is, E = ±√(p²c² + m²c⁴). For p → 0 the positive branch becomes mc² + p²/2m − p⁴/8m³c² + … (rest energy + the Schrödinger kinetic energy + the mass-velocity correction), connecting correctly to the nonrelativistic theory. The problem is the negative branch: there is energy without limit below −mc². For one-particle quantum mechanics this means that “there is no ground state,” and this is later understood in terms of antiparticles (positrons).

2.2 What “negative probability” really means

In the Schrödinger equation |ψ|² is the probability density, and ∂|ψ|²/∂t + ∇·j = 0 follows from the equation. A conservation law also follows for the KG equation, but because it is second order in time the conserved density is not |ψ|² but

(4) ρ=iℏ2mc2 (ψ*∂ψ∂t−ψ∂ψ*∂t) ⟹ for a plane waveρ=Emc2|ψ|2

which becomes ρ < 0 for negative-energy solutions (|ψ|² itself is of course positive; please read the “ψ*ψ < 0” of slide 30 as referring to this conserved density ρ). That ρ cannot be interpreted as a probability density is the reason why, in 1926, many people including Schrödinger abandoned the KG equation. Pauli and Weisskopf (1934) reinterpreted ρ as a “charge density” (+ for particles, − for antiparticles) and revived the KG equation as the correct equation for spin-0 particles (such as π mesons). Feshbach and Villars (1958) rewrote it as a first-order equation with two components (a particle component φ and an antiparticle component χ). If you choose “π+ meson (spin 0)” in Simulator ①, the weights of these two components are displayed.

It cannot be used for the electron: there is no room for spin in the KG equation, and applied to the hydrogen atom it gives a fine structure that disagrees with experiment (this is the calculation Schrödinger tried first and discarded). The electron, with spin 1/2, requires the Dirac equation.

3. The Dirac Equation — Factorization Demands 4×4 Matrices, and Spin and Antiparticles Emerge

3.1 Taking the square root “with matrices”

Dirac (1928) wanted to keep a Schrödinger-type equation that is first order in time, and assumed the linear expression E = cα·p + βmc² (slide 31). For its square to agree with E² = p²c² + m²c⁴ of Eq. (3), we need

(5) (cα·p+βmc2)2 =p2c2+m2c4 ⟺ αiαj+αjαi=2δij, αiβ+βαi=0, β2=1

To eliminate the cross terms (such as pxpy), αx, αy, αz, and β must mutually anticommute, which is impossible for numbers — matrices are required. The smallest size of four mutually anticommuting matrices is 4×4 (for 2×2 matrices such as the Pauli matrices, at most three can mutually anticommute), and in the standard representation they are

(6) α=(0σσ0), β=(100−1), iℏ∂ψ∂t= (−iℏcα·∇+βmc2)ψ

where σ denotes the Pauli matrices from the lecture. Therefore ψ is a four-component column vector (a spinor), and the Dirac equation is a set of four coupled first-order partial differential equations. Each component also satisfies the KG equation (squaring Eq. (6) gives Eq. (3)), so the dispersion relation is the same as for KG, E = ±√(p²c² + m²c⁴). The difference lies in the structure of the eigenfunctions.

3.2 Plane-wave solutions: large and small components

Inserting ψ = u(p) ei(p·r − Et)/ħ and splitting u into the upper two components uA and the lower two components uB, Eq. (6) becomes

(7) (E−mc2)uA=cσ·puB, (E+mc2)uB=cσ·puA ⟹ u(p)= E+mc22E (χcσ·pE+mc2χ), |uB||uA|=pcE+mc2≈v2c

Here χ is a two-component spin state (spin up and spin down; the α, β used for the spin functions in the lecture, not the Dirac matrices α, β); the upper two components are the “large components,” the lower two are the “small components,” and their ratio is of order v/2c. In the nonrelativistic limit the small components vanish, leaving a two-component Pauli spinor, and we return to the Schrödinger equation plus spin. Relativistic computer codes (four-component Dirac–Kohn–Sham) are more than twice as heavy as nonrelativistic ones because these small components must also be expanded in basis functions. ▶ Simulator ①: Weight of the small component

3.3 Negative-energy solutions, the Dirac sea, and antiparticles

Two of the four independent solutions have E < 0. Dirac (1930) regarded the vacuum as a “sea” in which all negative-energy states are filled with electrons (by the Pauli exclusion principle, positive-energy electrons cannot fall into it), and proposed that a “hole” in the sea looks like a particle with positive charge and positive energy. This is the positron, which Anderson discovered in cosmic rays in 1932. For condensed-matter physicists the direct analogy is “the sea of electrons filling the valence band, and holes” (the “≒ electrons & holes” of slide 31). The counterpart of the semiconductor gap Eg is 2mc² = 1.022 MeV, which is the threshold for pair production (the statement in §6 of the elementary-particle page that “if an electron is confined to a region smaller than ƛC, pairs are produced” is also based on this number).

3.4 Why does spin appear?

With the Dirac Hamiltonian H = cα·p + βmc², the orbital angular momentum L = r×p is not conserved: [H, L] = −iħc α×p ≠ 0. What is conserved is

(8) J=L+ℏ2Σ, Σ=(σ00σ), [H,J]=0

where the added ħΣ/2 is the spin, with eigenvalues ±ħ/2. The spin that in nonrelativistic theory was “attached by hand to fit experiment” enters automatically once a Lorentz-symmetric first-order equation is constructed — this is the greatest achievement of the Dirac equation. Moreover, the density of the Dirac equation, ρ = ψ†ψ = Σ|ψi|², is positive definite, so there is no interpretational problem as in KG, and (apart from the treatment of the negative-energy solutions) it can be used as it is as one-particle quantum mechanics. That the electron trembles among its four components on the scale of ƛC = ħ/mc (Zitterbewegung) was discussed in §6.2 of the elementary-particle page.

4. Gauge Symmetry and Minimal Coupling — The Freedom of Phase Demands an Electromagnetic Field

4.1 Global phase and local phase

Changing the phase of the wave function uniformly everywhere, ψ → eiθψ (θ a constant), leaves every equation invariant (neither |ψ|² nor the current changes). But what about changing the phase from place to place and from time to time, ψ → eiθ(r,t)ψ? The derivative picks up the gradient of θ, so the form of the equation breaks down as it stands: ∇(eiθψ) = eiθ(∇ + i∇θ)ψ. If we introduce a “partner field” that cancels this extra term and let that field transform along with ψ, the equation becomes invariant again:

(9) ∇→∇−iqℏA, ∂∂t→∂∂t+iqℏφ, A→A+ℏq∇θ, φ→φ−ℏq∂θ∂t

This transformation of (φ, A) is exactly the familiar gauge transformation of electromagnetism (a change of potentials that leaves E and B unchanged). In other words, if we impose the requirement that “the phase may be chosen freely at each place” (local U(1) symmetry), an electromagnetic field must exist, and even the form of its coupling (minimal coupling) is fixed. This is the gauge principle (Weyl 1929), and all three forces of the Standard Model (electromagnetic, weak, and strong) are obtained by extending this argument to U(1), SU(2), and SU(3) (§7 of the Lie group page).

4.2 The Schrödinger equation was gauge invariant from the start

(10) iℏ∂ψ∂t= [(p−qA)22m+qφ]ψ, j=1mRe[ψ*(p−qA)ψ]

The (p − qA)²/2m used in condensed-matter physics for electrons in a magnetic field is precisely Eq. (9) applied to the Schrödinger equation. Landau levels, the Aharonov–Bohm effect, the London equation of superconductivity, and flux quantization are all consequences of the fact that “the phase of ψ and A have meaning only as a set,” and observables (|ψ|², the gauge-invariant current j, E, B) do not depend on the choice of gauge. Relativity does not demand gauge symmetry, but gauge symmetry comes into its own precisely in relativistic theories (next section).

4.3 Putting minimal coupling into the Dirac equation gives g = 2 and the spin–orbit interaction

Substitute p → π ≡ p − qA and E → E − qφ into Eq. (6), and eliminate the small components as in Eq. (7). For a nonrelativistic energy E = mc² + ε (ε ≪ mc²), uB ≈ σ·π uA/2mc, so the equation for the large components is

(11) εuA= [(σ·π)22m+qφ]uA = [π22m−qℏ2mσ·B+qφ]uA ⟹ μ=gq2mS,g=2

(the Pauli equation). From the Pauli-matrix identity (σ·π)² = π² − qħσ·B, the spin magnetic-moment term comes out including its coefficient — g = 2, which in nonrelativistic theory could only be supplied as an experimental value, is derived (the radiative correction of QED, g = 2.0023…, is treated in §6.5 of the elementary-particle page). Expanding further to order 1/c² (the Foldy–Wouthuysen transformation) gives

(12) H≈mc2+π22m+qφ −qℏ2mσ·B −π48m3c2mass-velocity +12m2c21rdVdrL·Sspin–orbit +ℏ28m2c2∇2VDarwin

(with V = qφ for a central force), in which three relativistic correction terms appear. The mass-velocity term is the third term of the expansion in §2.1, the spin–orbit term is automatically correct even down to the Thomas-precession factor of 1/2, and the Darwin term is the shift of s levels due to the electron being smeared over a range ƛC (Eq. (20) of the elementary-particle page). The fine structure of hydrogen is fully explained by these three terms.

4.4 Connection to condensed-matter physics and materials computation

5. A Particle in a Box — The Relativistic Infinite Well

5.1 Schrödinger: ψ = 0 at the wall

For a one-dimensional infinite well of width L, ψn = √(2/L) sin(nπx/L) and En = ħ²π²n²/2mL². Eq. (2.33) of the doctoral thesis is the three-dimensional version of this, and is the starting point of the free-electron gas.

5.2 Klein–Gordon: same ψ, different E (and how to make the “wall”)

The KG equation is second order in space, so the same “ψ = 0 at the wall” as for Schrödinger can be used, giving kn = nπ/L and En = √(ħ²c²kn² + m²c⁴). However, one must be careful about how the “wall” is made. For the nonrelativistic infinite well it is fine to think of an electrostatic potential V → ∞, but in relativity a high wall of electrostatic potential does not confine the particle (Klein’s paradox, 1929). An electron incident on a step of height V0 connects, when its energy beyond the step becomes E − V0 < −mc², to a propagating solution in the negative-energy band and is transmitted without decay:

(13) V0>E+mc2 ⟹ in the wall (E−V0)2−m2c4>0 (real k: transmission, pair production)

Seen from field theory, the electric field of the wall creates electron–positron pairs out of the vacuum (Sauter 1931: the critical field m²c³/eħ = 1.3×1018 V/m is the field that gives mc² over a distance ƛC). Therefore a relativistic “box” is made by sending the mass term, which is a Lorentz scalar, to infinity outside (m(x) = M → ∞) [Alberto, Fiolhais, Gil 1996; Alberto, Das, Vagenas 2017]. The figure at the lower right of Simulator ② draws the overlap of the energy bands underlying Klein’s paradox.

5.3 Dirac: a first-order equation does not allow “ψ = 0” — the MIT boundary condition

In one dimension 2×2 Dirac matrices suffice; taking α = σx and β = σz, for ψ = (u, v) we have

(14) (E−mc2)u=−iℏcdvdx, (E+mc2)v=−iℏcdudx

Putting u = A sin kx + B cos kx gives v = −iκ(A cos kx − B sin kx), κ = ħck/(E + mc²). If we now impose “u = v = 0 at the wall,” all the solutions vanish (in a first-order equation the two components cannot both be set to 0 at the same time). Instead, we impose “zero flow through the wall” at the wall, that is, ψ̄ψ = |u|² − |v|² = 0 (the MIT bag boundary condition, which coincides with the limit of a mass wall M → ∞), and

(15) v(0)=−iu(0), v(L)=+iu(L) ⟹ tan(kL)=−2κ1−κ2=−ℏkmc

becomes the quantization condition (we used κ² = (E − mc²)/(E + mc²)). For L ≫ ƛC the right-hand side → 0 and kL → nπ (Schrödinger), and for L ≪ ƛC the right-hand side → −∞ and kL → (n − ½)π. The levels lie below those of KG (same mass, same box), and the density |u|² + |v|² at the wall is 2κ²|A|² ≠ 0 — “the particle is present on the wall” is a characteristic of relativistic quantum mechanics. ▶ Simulator ②: Particle in a box

5.4 The three-dimensional box and the counting of states — toward the free-electron gas

In a three-dimensional box, the wave numbers allowed for KG (and for Dirac with L ≫ ƛC) are the same as for Schrödinger, ki = niπ/L. Therefore the way of counting states in k space (the “sphere in quantum-number space” of Eq. 2.34 of the thesis) does not change in relativity; only the energy E(k) of each state changes. The Fermi wave number kF = (3π²n)1/3 is a quantity determined purely by “the number of electrons,” so it too stays as it is in relativity, and only the Fermi energy changes. This is the whole content of the next section.

6. The Relativistic Free-Electron Gas — From Metals to White Dwarfs

6.1 Density of states

The number of states with momentum magnitude up to p, including the spin factor 2, is N(p) = 2·(4πp³/3)·V/(2πħ)³ = Vp³/3π²ħ³. This is the same in relativity. Differentiating with respect to the kinetic energy E = √(p²c² + m²c⁴) − mc², and using dp/dE = (E + mc²)/pc², we obtain

(16) g(E)=1VdNdE =(E+mc2)E2+2Emc2π2ℏ3c3 → {12π2(2mℏ2)3/2E(E ≪ mc², Eq. 2.35 of the thesis)E2π2ℏ3c3(E ≫ mc², same as for photons)

√E at low energies and E² at high energies — for the same spacing in k, in relativity E grows more slowly (approaching E ∝ k), so there are more states per unit energy.

6.2 The Fermi energy and the relativity parameter xF

(17) kF=(3π2n)1/3, xF≡ℏkFmc=(3π2n)1/3λ¯C, EF=mc2(1+xF2−1) ≈ℏ2kF22m(1−xF24)

The density at which xF = 1 (pF = mc) is n = 1/(3π²ƛC³) = 5.9×1029 cm−3, which for a material with μe = 2 is 2×106 g/cm³. Compared with metals,

Systemn [cm−3]kF [Å−1]xFEFRelativistic correction −xF²/4
Na2.65×10220.923.6×10−33.24 eV−3×10−6
Cu8.47×10221.365.2×10−37.0 eV−7×10−6
Al1.81×10231.756.8×10−311.7 eV−1×10−5
Near a gold nucleus (1s density 2Z³/π)2×1030397 (210 a.u.)1.50.41 MeV(cannot be expanded)
Center of a white dwarf1030–1032—1–60.2–3 MeV(ultrarelativistic)

For conduction electrons in metals the correction is 10−5: this is why condensed-matter physics can get by with the Schrödinger equation. The inner shells of heavy atoms (kF ≈ 2.66Z a.u. [Paquier & Toulouse 2018]) and white dwarfs, on the other hand, lie in the “relativistic” regime of the same free-electron-gas formulas. ▶ Simulator ③: Free-electron gas

6.3 Kinetic energy and pressure: Chandrasekhar’s formulas

The kinetic energy per electron is obtained by integrating g(E) up to the Fermi level and dividing by n, and the pressure is the momentum flux, P = (1/3)∫v p dn (v = pc²/E). Both take closed forms in x = p/mc [Chandrasekhar 1931, 1935]:

(18) TN=3mc28xF3 [xF(1+2xF2)1+xF2−sinh−1xF]−mc2 →35EF(nonrelativistic, Eq. 2.38 of the thesis) P=m4c524π2ℏ3 [xF(2xF2−3)1+xF2+3sinh−1xF] → {25nEF∝n5/3(xF≪1)14npFc∝n4/3(xF≫1)

The pressure is “particle flux × momentum,” so P ∼ n vFpF. Nonrelativistically vF ∝ pF ∝ n1/3, hence n5/3; relativistically vF saturates at c, hence n4/3 — relativity makes the degeneracy pressure “softer.” This is the essence of the Chandrasekhar limit that follows.

6.4 White dwarfs and the Chandrasekhar limit

A white dwarf is an object whose core (carbon and oxygen, μe = 2), left over after a star has finished nuclear fusion, supports its own weight by electron degeneracy pressure. Solving the hydrostatic equilibrium dP/dr = −Gmρ/r² with P = Kργ (a polytrope), the mass varies with the central density as M ∝ ρc(3γ−4)/2. For γ = 5/3, M ∝ ρc1/2, and heavier stars shrink and become denser (the radius is R ∝ M−1/3). For γ = 4/3, however, the exponent becomes 0: the mass is fixed at a single value independent of the central density. That value is

(19) MCh=4π·2.018 [ℏc(3π2)1/34πG(μemu)4/3]3/2 ≈5.83μe2M⊙=1.46M⊙(μe=2)

Here 2.018 is the constant −ξ²θ′(ξ1) of the Lane–Emden function of index 3; the customary value, “1.44 M⊙,” includes corrections for finite temperature, inverse β decay, and general relativity. Landau’s (1932) estimate is more intuitive: if N ultrarelativistic electrons are confined in a radius R, the kinetic energy is N pFc ∝ N4/3ħc/R and the gravitational energy is −GM²/R ∝ −N²Gmu²/R. Both go as 1/R, so if N4/3ħc > N²Gmu² the star expands, returns to the nonrelativistic regime, and is stable; otherwise it collapses without limit. The boundary is N ∼ (ħc/Gmu²)3/2 ≈ 2×1057 electrons, M ≈ 1–2 M⊙. ħc/Gmu² = 1/αG is the inverse of the “gravitational coupling constant” in §7 of the elementary-particle page, and the maximum mass of a star is determined by quantum mechanics (degeneracy pressure), relativity, and gravity alone, without using the electromagnetic force or the nuclear force.

Central density ρc [g/cm³]10610710810910101011
Central xF0.801.73.78.01737
Mass M [M⊙]0.400.821.181.371.431.45
Radius R [km]11,0007,1004,3002,5001,300660

(Values obtained in Simulator ③ by numerically integrating the equation of state of Eq. (18) together with hydrostatic equilibrium under Newtonian gravity. Sirius B, with M = 1.02 M⊙ and R = 5,800 km, lies on this curve.) The more the mass increases, the more the radius shrinks, and as the center becomes relativistic the mass approaches 1.46 M⊙ asymptotically — the conclusion of this section is that merely “putting relativity into the free-electron gas” lets us see the threshold of supernovae and neutron stars. ▶ Simulator ③: M–R relation of white dwarfs

6.5 Making the Thomas–Fermi kinetic-energy functional relativistic

Eqs. (2.39)–(2.40) of the thesis were obtained by evaluating “(3/5)EF per electron” at the density of each point and integrating. In relativity, using T/N of Eq. (18), we obtain

(20) TTF[ρ]= ∫ρ(r)t(xF(ρ))d3r, xF(ρ)=(3π2ρ)1/3λ¯C → {310(3π2)2/3∫ρ5/3(nonrel., Eq. 2.40)34(3π2)1/3ℏc∫ρ4/3(ultrarelativistic)

Here ρ5/3 turns into ρ4/3 — this one line is the whole of the relativistic Thomas–Fermi atom of §8 (where the local xF exceeds 1 near the nucleus).

7. Relativistic Exchange Energy — The Relativistic Version of Dirac Exchange and the Sign Reversal

7.1 The nonrelativistic Dirac exchange (Eqs. 2.53–2.54 of the thesis)

The exchange energy computed with a plane-wave Slater determinant, using the Fock–Dirac density matrix ρ(r1,r2) = 3ρ (sin t − t cos t)/t³ (t = kFr12), was

(21) εxNR=−34πkF[hartree] ⟺ Ex[ρ]=−34(3π)1/3∫ρ4/3d3r

(the value per electron and the form of the functional; the same as the LDA exchange of Eq. 2.63).

7.2 Doing the same calculation with Dirac spinors

In relativity the plane waves become the spinors of Eq. (7), and the spinor overlap |u†(k1)u(k2)|² appearing in the exchange integral becomes smaller than 1 for k1, k2 pointing in different directions (it has the form [E1E2 + ħ²k1·k2c² + m²c⁴]/2E1E2). As a result the Coulomb exchange is reduced. In addition, moving charges also interact magnetically (the Breit interaction = the 1/c² approximation to the exchange of transverse photons), and its exchange term is positive. With β = xF and c̃ = 1/β [MacDonald & Vosko 1979; Paquier & Toulouse 2018]:

(22) εxC=εxNR [56+c~23+231+c~2sinh−11c~ −13(1+c~2)2ln(1+1c~2) −12f2], f≡1+c~2−c~2sinh−11c~ εxB=−εxNR [1−2(1+c~2)(1−c~2ln1+c~2c~2)+2f2], εxQED=εxNR[1−32f2]

The last expression is the relativistic LDA exchange of MacDonald and Vosko (the total exchange obtained with the QED photon propagator); many terms of the Coulomb and transverse (Breit + retardation) parts cancel, leaving a simple form. The limits and numerical values are

β = xF0.10.5122.535→∞
Coulomb only εxC/εxNR0.9990.9760.9320.8680.8500.8380.814(1 + ln 4)/3 = 0.795
Breit εxB/εxNR−0.006−0.120−0.340−0.652−0.739−0.798−0.911−1
Coulomb + Breit0.9930.8570.5910.2150.1110.039−0.097−0.205
QED (MacDonald–Vosko)0.9930.8550.5740.1400.007−0.089−0.290−1/2

At β ≈ 2.5 (kF ≈ 350 a.u., n ≈ 1031 cm−3) the exchange energy becomes zero, and beyond that it becomes positive. If Coulomb exchange is “the energy gained because electrons of the same spin avoid one another (the Fermi hole),” then Breit exchange is “the loss incurred when the exchange hole removes the magnetic attraction between parallel currents running side by side,” and in the ultrarelativistic limit the latter wins [Jancovici 1962]. The inner shells of heavy atoms (kF ≈ 2.66Z a.u., 210 a.u. for Hg, β ≈ 1.5) lie in just the region where this factor is about 0.3, and a self-consistent treatment of transverse exchange affects the total energy for Z ≳ 50 [Engel 2002]. ▶ Simulator ③: Exchange factor Φ(β)

7.3 Relativistic LDA and the Dirac–Kohn–Sham equation

(23) [cα·p+βmc2+vext+vH+vxc]ψi=εiψi, ExRLDA[ρ]=∫ρεxNR(ρ)Φ(β(ρ))d3r

In relativistic density functional theory [Rajagopal 1978; MacDonald & Vosko 1979], the kinetic-energy operator of the Kohn–Sham equation is replaced by the Dirac operator (four-component ψi), and the exchange-correlation functional is approximated by εxc of the relativistic electron gas (exchange from Eq. (22), correlation from perturbation theory or quantum Monte Carlo [Kenny et al. 1996]). The LDA of Eqs. (2.62)–(2.66) of the thesis multiplied by the factor Φ of Eq. (22) is the RLDA. Most of the relativistic effects come from the kinetic-energy operator (that is, orbital contraction and spin–orbit), and the relativistic correction to exchange-correlation is about 0.1% of the total energy of heavy elements, but it cannot be neglected for inner-shell levels and X-ray transitions.

8. The Relativistic Thomas–Fermi Model — The Vallarta–Rosen Equation

8.1 The nonrelativistic Thomas–Fermi model (summary of §2.1.6 of the thesis)

Regarding the electrons at each point as a local free-electron gas, we impose the condition that the chemical potential is constant (0 for a neutral atom): (3π²ρ)2/3/2 − φ(r) = 0 (Hartree units, Eq. 2.43) gives ρ = (2φ)3/2/3π² (Eq. 2.44), and together with the Poisson equation, with φ = Zχ/r and x = r/b, we obtain

(24) d2χdx2=χ3/2x, b=(9π2128)1/3Z−1/3a0=0.8853Z−1/3a0, χ(0)=1,χ(∞)=0,χ′(0)=−1.588071

This is Eq. (2.47). The total energy is ETF = (3/7)(128/9π²)1/3Z7/3χ′(0) = −0.7687Z7/3 hartree (Eq. 2.50), and Table 2.5 of the thesis shows that E0 = −0.7687Z7/3 + 0.5Z² − 0.2699Z5/3 (Eq. 2.55), which includes Schwinger’s correction, agrees with Hartree–Fock to within 0.1%.

8.2 The relativistic equilibrium condition and the Vallarta–Rosen equation

If the Fermi energy of the local electron gas is made relativistic, the equilibrium condition becomes

(25) pF2c2+m2c4−mc2−eφ(r)=0 ⟹ pFc=(eφ)2+2mc2eφ ρ=pF33π2ℏ3=ρNR(1+eφ2mc2)3/2

that is, the density acquires a factor (1 + eφ/2mc²)3/2 (the same as the increase in the density of states in Eq. (16)). In the variables of the thesis, writing eφ = Ebχ/x and Eb = Ze²/4πε0b = 30.7Z4/3 eV, the Poisson equation becomes

(26) d2χdx2=χ3/2x (1+εχx)3/2, ε=Eb2mc2=Zα2λ¯Cb=3.0×10−5Z4/3

This is the Vallarta–Rosen equation (1932). Since ε is small (0.010 for Hg and 0.012 for U), εχ/x ≳ 1 for x ≲ ε, that is, inside r ≲ εb ≈ ZαƛC/2 (≈ 100 fm for Hg), eφ exceeds 2mc² and the electron gas becomes fully relativistic (the radius rrel at which pF = mc, a somewhat looser criterion, that is, eφ = (√2 − 1)mc², lies about 5 times farther out, at ≈ 540 fm, and the table in §8.4 uses this one). Rotondo et al. write the same equation in units of the pion Compton wavelength; the only difference is that χ is defined by eφ = ħcχ/r [Rotondo et al. 2011, Eq. (23)].

8.3 For a point-charge nucleus no solution exists

If χ → 1 near the origin, eφ ≈ Ze²/4πε0r → ∞ and the electron gas becomes ultrarelativistic, so

(27) ρ≈(eφ)33π2(ℏc)3∝1r3 ⟹ N(<r)=∫4πr2ρdr∝lnr→−∞

the electron number diverges logarithmically at the origin, and χ also diverges as χ ∼ −ε3/2 ln x (from χ″ ≈ ε3/2/x²), so χ(0) = 1 cannot be imposed. “A point-charge nucleus + a relativistic electron gas” is inconsistent, and the relativistic TF model can be defined only when the size of the nucleus is included [Vallarta & Rosen 1932; Rotondo et al. 2011]. For a uniformly charged nucleus of radius Rc = 1.2A1/3 fm, a term for the nuclear charge density, −(3x/xc³)θ(xc − x) (xc = Rc/b), is added to the right-hand side of Eq. (26), and the boundary conditions are χ(0) = 0 and χ(∞) = 0. The slope at the origin, χ′(0) ≈ 1/xc (χ rises from 0 to ≈ 1 inside the nucleus), is determined by the shooting method. ▶ Simulator ④: The relativistic Thomas–Fermi atom

8.4 What changes: numbers for Hg (Z = 80)

r11 fm (x = 10−3)110 fm (10−2)540 fm (≈ rrel)1.1 pm (0.1)11 pm (1)
χ (nonrelativistic, point nucleus)0.99850.9850.9370.8830.420
χ (relativistic, finite nucleus)0.9950.9780.9250.8670.415
N(<r)/Z (nonrelativistic)2.0×10−56.7×10−46.8×10−31.8×10−20.306
N(<r)/Z (relativistic)1.2×10−35.0×10−31.5×10−23.0×10−20.315

Although the potential (χ) hardly changes, the number of electrons right next to the nucleus increases by a factor of several to several tens — this is exactly the conclusion Vallarta and Rosen reported for Hg. However, rrel ≈ 0.5 pm (the radius at which the electron gas becomes relativistic) is comparable to the Bohr K-shell radius a0/Z = 0.66 pm, TF places about one electron there, and in reality there are two 1s electrons. That is, the region where relativity matters is the region where the TF premise of a “local electron gas” is at its worst.

8.5 The magnitude of the relativistic corrections, and the limits of TF

A guide to the relativistic correction of the total energy is the sum, shell by shell, of corrections such as the Dirac 1s-level correction −(Zα)²Z²/8 hartree, and for heavy elements it amounts to a few percent of the total energy. For Hg, for example, against the nonrelativistic Hartree–Fock value of −18,409 hartree, Dirac–Fock (Coulomb) gives −19,649 hartree, and including the Breit interaction −19,626 hartree [Engel 2002, Table 1]: The relativistic correction of −1,240 hartree is of the same order as the correction terms in Eq. (2.55) of the thesis, 0.5Z² = 3,200 hartree and 0.2699Z5/3 = 400 hartree. In other words, for heavy elements even the TFD energy, matched to 0.1% by Schwinger’s correction, is off by 6% unless relativity is included. Relativistic TF takes this correction in as a change of the density, but because the local-electron-gas approximation is poor in the 1s region, it does not become quantitative. That is why modern relativistic atomic and solid-state calculations solve the Dirac–Kohn–Sham equation of Eq. (23) (or the two-component scalar relativity + spin–orbit), and the knowledge of the relativistic electron gas is used on the exchange-correlation functional side (§7.3). Making the TF model — “the origin of density functional theory” — relativistic, including its limits, teaches us why relativistic DFT has the form it has today.

9. Correspondence Table: Nonrelativistic and Relativistic

Schrödinger (nonrelativistic)Klein–Gordon (spin 0)Dirac (spin 1/2)
Starting pointE = p²/2mE² = p²c² + m²c⁴E = cα·p + βmc²
Order of derivatives (time / space)1 / 22 / 21 / 1
SymmetryGalileiLorentzLorentz
Components of the wave function1 (spin added separately)1 (2 in the FV representation)4 (spin × particle/antiparticle)
Dispersion relationparabola± hyperbola± hyperbola
Conserved density|ψ|² ≥ 0 (probability)(iħ/2mc²)(ψ*∂tψ − c.c.) (charge)ψ†ψ ≥ 0 (probability)
Spin and g factoradded by handnoneautomatic (g = 2)
Boundary condition of the boxψ = 0ψ = 0 (mass wall)ψ̄ψ = 0 (MIT), tan kL = −ħk/mc
Coupling to the electromagnetic field(p − qA)²/2m + qφsame (KG type)cα·(p − qA) + qφ → Pauli, SO, Darwin
Density of states (electron gas)∝ √E(E + mc²)√(E² + 2Emc²)/π²ħ³c³ → ∝ E²
Fermi energyħ²kF²/2mmc²(√(1 + xF²) − 1), xF = ħkF/mc
Pressure(2/5)nEF ∝ n5/3Chandrasekhar’s formula → (1/4)npFc ∝ n4/3 (limiting mass)
Exchange energy−(3/4π)kF—×Φ(β): 1 → 0 (β ≈ 2.5) → −1/2
TF equationχ″ = χ3/2/√xχ″ = (χ3/2/√x)(1 + εχ/x)3/2, finite nucleus required

10. Implementation Notes for the Simulator

11. References

  1. Yasuhide Mochizuki, Computational Materials Science, Lecture 7 (lecture slides 29–31, in Japanese: time–space symmetry in the Schrödinger equation, the Klein–Gordon and Dirac equations); doctoral thesis, Chapter 2, §2.1.6–2.1.8 (the Thomas–Fermi model, the Thomas–Fermi–Dirac model, and the Kohn–Sham equation).
  2. P. A. M. Dirac, The Quantum Theory of the Electron, Proc. R. Soc. A 117, 610 (1928); A Theory of Electrons and Protons, Proc. R. Soc. A 126, 360 (1930).
  3. O. Klein, Z. Phys. 37, 895 (1926); W. Gordon, Z. Phys. 40, 117 (1926); W. Pauli and V. Weisskopf, Helv. Phys. Acta 7, 709 (1934); H. Feshbach and F. Villars, Rev. Mod. Phys. 30, 24 (1958).
  4. H. Weyl, Elektron und Gravitation. I, Z. Phys. 56, 330 (1929). The gauge principle.
  5. L. L. Foldy and S. A. Wouthuysen, Phys. Rev. 78, 29 (1950). The nonrelativistic expansion (mass-velocity, Darwin, spin–orbit).
  6. O. Klein, Z. Phys. 53, 157 (1929); F. Sauter, Z. Phys. 69, 742 (1931). Klein’s paradox.
  7. P. Alberto, C. Fiolhais, and V. M. S. Gil, Relativistic particle in a box, Eur. J. Phys. 17, 19 (1996); P. Alberto, S. Das, and E. C. Vagenas, Phys. Lett. A 375, 1436 (2011); Relativistic particle in a box: Klein–Gordon versus Dirac equations, arXiv:1711.06313 (2017).
  8. S. Chandrasekhar, The Maximum Mass of Ideal White Dwarfs, Astrophys. J. 74, 81 (1931); Mon. Not. R. Astron. Soc. 95, 207 (1935); L. D. Landau, Phys. Z. Sowjetunion 1, 285 (1932).
  9. L. H. Thomas, Proc. Camb. Phil. Soc. 23, 542 (1927); E. Fermi, Rend. Accad. Naz. Lincei 6, 602 (1927); P. A. M. Dirac, Note on Exchange Phenomena in the Thomas Atom, Proc. Camb. Phil. Soc. 26, 376 (1930).
  10. M. S. Vallarta and N. Rosen, The Relativistic Thomas–Fermi Atom, Phys. Rev. 41, 708 (1932).
  11. M. Rotondo, J. A. Rueda, R. Ruffini, and S.-S. Xue, Relativistic Thomas–Fermi treatment of compressed atoms and compressed nuclear matter cores of stellar dimensions, Phys. Rev. C 83, 045805 (2011); arXiv:0911.4622.
  12. A. K. Rajagopal, J. Phys. C 11, L943 (1978); A. H. MacDonald and S. H. Vosko, A relativistic density functional formalism, J. Phys. C 12, 2977 (1979); M. V. Ramana and A. K. Rajagopal, Adv. Chem. Phys. 54, 231 (1983). Relativistic DFT and the exchange energy.
  13. B. Jancovici, On the relativistic degenerate electron gas, Nuovo Cimento 25, 428 (1962).
  14. E. Engel, Relativistic Density Functional Theory: Foundations and Basic Formalism, in Relativistic Electronic Structure Theory, Part 1 (ed. P. Schwerdtfeger, Elsevier, 2002); E. Engel and R. M. Dreizler, Density Functional Theory: An Advanced Course (Springer, 2011).
  15. J. Paquier and J. Toulouse, Four-component relativistic range-separated density-functional theory: Short-range exchange local-density approximation, J. Chem. Phys. 149, 174110 (2018); arXiv:1812.02024. The source of Eq. (22).
  16. S. D. Kenny, G. Rajagopal, R. J. Needs, W.-K. Leung, and W. M. C. Foulkes, Quantum Monte Carlo Calculations of the Energy of the Relativistic Homogeneous Electron Gas, Phys. Rev. Lett. 77, 1099 (1996).
  17. P. Pyykkö, Relativistic Effects in Structural Chemistry, Chem. Rev. 88, 563 (1988).
  18. N. W. Ashcroft and N. D. Mermin, Solid State Physics (Saunders, 1976), Table 1.1 (free-electron densities of metals).
  19. Textbooks: W. Greiner, Relativistic Quantum Mechanics: Wave Equations (3rd ed., Springer); J. J. Sakurai, Advanced Quantum Mechanics (Addison-Wesley); S. L. Shapiro and S. A. Teukolsky, Black Holes, White Dwarfs, and Neutron Stars (Wiley, 1983) — the standard derivation of the Chandrasekhar model.

Mochizuki Research Group, Tokyo University of Science. As a sequel to Lecture 7 of Computational Materials Science (the Klein–Gordon and Dirac equations) and Chapter 2 of the doctoral thesis (the Thomas–Fermi and Thomas–Fermi–Dirac models), this page summarizes, with derivations, what changes when the most basic examples of nonrelativistic quantum mechanics — the free electron, the infinite well, the free-electron gas, and the TF atom — are made relativistic. For numerical experiments, go to the simulator (Relativistic-QM-simulator-en.html). Related: Introduction to Elementary Particles, Lie Groups and Lie Algebras, The Hydrogen Atom. The formulas are typeset in the browser-standard MathML (there are no external libraries or network communication whatsoever).