Why the Energy Changes Depending on How Spins Are Arranged

This page writes out, step by step, the derivations from Lecture 7 (Slater determinants and spin) and Lecture 9 (Coulomb and exchange integrals). Expectation value of a Slater determinant → U and J → the accounting rule for filling → Hund’s rule and high spin/low spin → ligand field (point-charge model) → multiplets (CrystOD): a single straight path.

0. What Is the Problem? — The 2p² Terms of the Carbon Atom

The 2p² configuration of the C atom gives the two electrons a choice: their spins can line up (S = 1, ³P) or cancel each other (S = 0, ¹D and ¹S). The experimental values (Table 7.6 in Chapter 7 of the textbook “Introduction to Computational Materials Science,” in Japanese) show that the spin-aligned ³P is the ground state, while ¹D lies as much as 1.26 eV higher and ¹S as much as 2.68 eV higher. And this is despite the fact that the one-electron levels are the same in both cases. What makes this difference is the exchange integral J in the electron–electron repulsion, which is the star of this page.

1. The Conclusion First (the 30-Second Version)

  1. Taking the expectation value of the electron–electron repulsion 1/r12 with an antisymmetrized wavefunction (a Slater determinant) gives a result of the form Coulomb integral U − exchange integral J.
  2. The exchange integral survives only for electron pairs with the same spin (for opposite spins it vanishes by the spin orthogonality ⟨α|β⟩ = 0). Since J > 0, each same-spin pair is stabilized, with an energy shift of −J.
  3. So in degenerate orbitals it pays to enter “different orbitals with the same spin”—Hund’s rule.
  4. When a ligand field Δ is applied, the gain from −J and the loss from Δ compete, and the ground state switches between high spin ⇔ low spin.
  5. The pattern of the splitting (which orbitals remain degenerate) is fixed by symmetry, that is, by group theory (crystod-group --ligand-field), whereas the ordering and the size of the levels are decided by physics (the point-charge model, etc.).
  6. The true eigenstates beyond a single determinant are the multiplets (term symbols). The decomposition for each configuration can be computed with crystod-group --multiplet.

2. The Expectation Value of a Slater Determinant — Where U and J Come From

2.1 Spin orbitals and three configurations (Lecture 9)

From the spatial orbitals φ1 (1s), φ2 (2s) and the spins α, β, we form the spin orbitals χ1 = φ1α, χ̄1 = φ1β, … and write the antisymmetric two-electron wavefunction as a Slater determinant. The Hamiltonian is Ĥ = h1 + h2 + 1/r12 (h is the one-electron part). Taking the expectation value, the two-electron integral always takes the form

⟨ψ| 1r12 |ψ⟩ = ⟨χaχb| 1r12 |χaχb⟩ Coulomb integral U − ⟨χaχb| 1r12 |χbχa⟩ exchange integral J … ①

(The two terms come from expanding the determinant; the coordinates are ordered as ⟨x1x2|…|x1x2⟩.) The second term is the exchange term: the orbitals occupied by electrons 1 and 2 are swapped between the bra and the ket. This is where the orthogonality of the spins comes in. If χa and χb have opposite spins, the spin part of the exchange term contains ⟨α|β⟩ = 0, and the term vanishes entirely. For the same spin, ⟨α|α⟩ = 1 and it survives. The result (Lecture 9, slide 17):

ConfigurationTotal energyComment
1s↑1s↓ (ground)2h₁₁ + U₁₁In the same orbital the spins must be opposite (Pauli). The exchange term vanishes through ⟨α|β⟩ = 0, leaving only U
1s↑2s↓ (excited, antiparallel)h₁₁ + h₂₂ + U₁₂J vanishes by spin orthogonality
1s↑2s↑ (excited, parallel)h₁₁ + h₂₂ + U₁₂ − J₁₂Lowered by the term −J₁₂: the triplet is favored
J is the gain from same-spin electrons avoiding each other Two electrons with the same spin cannot come to the same place because of antisymmetry (the Fermi hole). Because they physically move apart, the Coulomb repulsion they pay comes out lower, by the term −J—that is what the exchange integral really is. J12 = ∬φ1*(r₁)φ2(r₁)·(1/r12)·φ2*(r₂)φ1(r₂)dV₁dV₂ takes the form of the “self-repulsion of the overlap density φ1φ2,” so it is always positive.

2.2 Measuring U and J directly with PySCF (Lecture 9, slides 21–22)

6-31G basis [eV]TotalKineticNuclear attractionHartree (U)Exchange (J)
H atom (1 electron)−13.55813.874−27.4328.554−8.554
He atom (2 electrons)−77.69377.697−183.33455.887−27.944

The highlight is the H atom: although there is only one electron, both a Hartree term and an exchange term appear, and they cancel exactly (the “Exchange (J)” column of the table is the exchange energy, that is, the negative value corresponding to −J in the main text). The Hartree term is an accounting scheme that even counts the repulsion with the cloud the electron itself creates, so the exchange term refunds this self-repulsion exactly (cancellation of the self-interaction). For He, U − J = 55.89 − 27.94 = 27.94 eV is the net electron–electron repulsion in Hartree–Fock. To compare on the same footing, look at how much the total energy rises above its value with the electron–electron repulsion switched off (for hydrogen-like He, 2 × (−Z²/2) = −4 hartrees = −108.85 eV): Hartree–Fock 31.16 eV, exact value 29.84 eV, variational 31.36 eV, and first-order perturbation 34.02 eV (Lecture 9, slide 22).

3. The Accounting Rule for Filling with Electrons

Applying ① to every pair of electrons, we can read off the energy of any configuration. Let εi be the one-electron term of spin orbital i (= the level in the ligand field); then

E= ∑i εi + ∑i<j ( Uij − Jij δσiσj ) … ②

This can be checked with the seven configurations of the exercise (Lecture 9, Szabo–Ostlund Problem 2.23). Use Fig. 1 to check your answers.

Fig. 1 | Total energies of the seven configurations (Szabo–Ostlund 2.23). Choose how levels 1 and 2 are filled, (a)–(g), and the picture of the configuration and the terms of the accounting ② appear. Check that −J₁₂ is added once for every same-spin pair.

In a degenerate shell, U is coarse-grained into two values Within the same shell (2p, 3d, 4f), the Uij in ② are lumped into two values. A pair entering the same orbital as ↑↓ gets U0, a pair entering different orbitals gets U′ = U0 − 2J, and only the same-spin pairs among the latter get a further −J—these three lines suffice to write the energy of every configuration. U′ = U0 − 2J is a relation that follows from the requirement that “the value does not change when the orbitals are re-chosen within the shell,” and it puts ② into a form that can be driven by just two parameters, U0 and J (Hubbard–Kanamori type). Fig. 2 below and tab ① (Fill with electrons) of the simulator build the total energy from these three lines.

4. Hund’s Rule and the Tug of War Between High Spin and Low Spin

In a degenerate shell (the 2p, 3d, and 4f of a free ion) the one-electron terms in ② are all the same, so the filling that maximizes the number of same-spin pairs = maximum S wins (Hund’s first rule). That ³P lies 1.26 eV below ¹D and 2.68 eV below ¹S in the 2p² of carbon (Section 0) is due to this −J (viewed term by term as in Section 6, the model with U′ = U0 − 2J gives ¹D − ³P = 2J and ¹S − ³P = 5J, so the differences are not a single J).

Once a ligand field Δ is applied, the story changes. For octahedral 3d⁴–3d⁷ it becomes a competition between “going up into eg to keep the spins parallel (high spin)” and “not paying Δ and forming pairs in t2g (low spin),” and the ground state switches depending on the size of Δ relative to the pair-formation cost (U, J). See Fig. 2.

Fig. 2 | Switching between high spin ⇔ low spin (3dN in an octahedron). The lowest energies of the maximum-S configuration (high spin) and of the configuration with t₂g filled as much as possible (low spin) are compared as functions of Δ. The crossing point Δc is where the spin crossover occurs (U₀ = 4 eV, J = 0.8 eV, t2g = −0.4Δ, eg = +0.6Δ).

5. Ligand Field — Deriving the Splitting from the Point-Charge Model

We expand the electrostatic potential of the ligands (anions) felt by an orbital of the central cation in a multipole expansion.

1|r−R| = ∑k rkRk+1 Pk (cos⁡γ) (r<R)

Only even k ≤ 2l contribute to the matrix elements within the same l (parity and the triangle condition). That is, p has only k = 2, d has k = 2, 4, and f has k = 2, 4, 6. In cubic symmetry (m-3m, -43m) the k = 2 term vanishes as well, and the splitting of d is determined by the single k = 4 term. Evaluating the matrix elements for an octahedron (six-coordinate) gives

Dq= qe02 ⟨r4⟩ 24πε0R5 , Et2g=−4Dq , Eeg=+6Dq

The center of gravity (3 × (−4Dq) + 2 × 6Dq = 0) is conserved. A way to remember the result is “orbitals pointing at the anions lose out”: the eg orbitals (z², x²−y²) pointing at the ligands on the axes go up, and the t2g orbitals (xy, yz, zx) that avoid the axes go down. In a tetrahedron the ligands are off the axes, so the order of the levels is inverted, and for the same ligands at the same distance R the point-charge model gives Δt = (4/9)Δo. For f orbitals, xyz (a2u), with lobes in eight directions, is the most stable, and a2u < t2u < t1u. In 4/mmm (elongation/compression) the k = 2 term comes back and the levels split into four (five for f), and the splitting of the eg degeneracy is the entrance to the Jahn–Teller effect. For numerical experiments, go to tab ② (Ligand field) of the simulator (the same point-charge model as on this page is running there).

⚠ Point charges are a model of “direction”; the magnitude is mostly a matter of covalency The measured 10Dq (1–3 eV for 3d oxides) has a large contribution from hybridization with the ligand orbitals, so the electrostatics of point charges alone gives too small a value. The pattern can be trusted because it is protected by symmetry, but the ratio 4/9 is the value of a point-charge model that assumes the same ligands at the same distance R (in real complexes both R and the covalency differ depending on the coordination, so deviations arise). In practice, the quantitative magnitude is left to spectroscopic data or first-principles calculations.

6. Multiplets — One Determinant Is Not the Whole Story

The accounting in ② gives the energy of “a single Slater determinant.” But there is more than one way to place electrons in a degenerate shell: for p² there are 6C2 = 15 ways (Lecture 7, p. 25). The true eigenstates are superpositions of these 15 determinants, and for a free ion they split into three multiplets, ³P (9 states) + ¹D (5 states) + ¹S (1 state), ordered ³P < ¹D < ¹S according to Hund’s rules. The same happens in a ligand field; for example, the octahedral (t2g)² is decomposed as

(T2g)2 = 3T1g + 1A1g + 1Eg + 1T2g (15 states)

CrystOD returns this decomposition and even the energies of the multiplets (in terms of the Racah parameters A, B, C):

Multiplets of (T2g)²Energy (Coulomb part)
³T1g (ground)A − 5B
¹Eg, ¹T2gA + B + 2C
¹A1gA + 10B + 5C

The ground term is the triplet ³T1g—Hund’s rule is at work here as well. The uniform U, J accounting of ② amounts to a coarse-graining that “averages” this multiplet structure (the difference between the triplet and the singlets appears, but not the structure of B and C, such as the difference between ¹Eg and ¹A1g). How the energy-level diagram of the multiplets changes with the ratio of Δ to B is organized in the Tanabe–Sugano diagrams, which are used to interpret the optical absorption of transition-metal compounds.

7. How to Calculate with CrystOD

All the group-theory data on this page and in the simulator were produced with CrystOD. You can reproduce them on your own machine:

8. Correspondence with the Lectures

References: lecture course “Computational Materials Science,” Lectures 7 and 9 / A. Szabo and N. S. Ostlund, Modern Quantum Chemistry (Japanese translation: Atarashii Ryoshi Kagaku), Problem 2.23 / Toyohiko Konno, Symmetry of Matter and Group Theory (in Japanese) / S. Sugano, Y. Tanabe, and H. Kamimura, Multiplets of Transition-Metal Ions in Crystals / group-theory calculations with CrystOD (crystod-group v0.4.4). The formulas are typeset with the browser’s standard MathML (no external libraries).