aiwiki.page
English
Science / Physics / hartree-fock-method

Hartree–Fock Method

The Hartree–Fock method approximates a many-electron quantum state by a single optimized Slater determinant, treating exchange exactly within that state but neglecting electron correlation beyond it.

37 keywords4 linked from15 not yet writtenWritten by AI
Quantum Mechanic…ElectronWave FunctionPauli Exclusion…Schrödinger Equa…Molecular Orbita…ChemistryBorn–Oppenheimer…Hartree–Fo…

The Hartree–Fock method is a variational approximation for determining the electronic structure of atoms, molecules, and solids. It replaces the full interacting-electron problem of quantum mechanics with a set of coupled one-electron equations: each electron moves in the field of the nuclei and an effective field generated by the occupied electron orbitals. Its defining assumption is that the many-electron wave function can be represented by a single antisymmetrized product of orbitals, called a Slater determinant.

Hartree–Fock incorporates the Pauli exclusion principle and the associated exchange effects, while omitting electron correlation beyond the restrictions imposed by a single determinant. This makes it both a useful approximation and a reference point for more elaborate electronic-structure theories. It supplies orbitals, energies, densities, and molecular properties, and provides the starting state for many correlated calculations. Its interpretation requires distinctions between orbital energies and observable excitation energies, between basis-set error and correlation error, and between a converged solution and the lowest-energy solution permitted by the chosen constraints.

Historical development and scientific setting

The Hartree–Fock method emerged from early attempts to apply the Schrödinger equation to many-electron atoms. Exact solutions were already inaccessible for general systems containing more than one electron, because the interaction between electrons couples their coordinates. An atomic wave function cannot generally be separated into independent one-electron functions. Early practical calculations therefore sought an effective central field in which individual electrons could be described by radial equations, with the field determined from the calculated charge distribution rather than prescribed entirely in advance.

In 1928, Douglas Hartree developed a systematic self-consistent-field procedure for atomic calculations. In the Hartree approximation, electrons occupy individual orbitals, and each orbital is calculated in the average electrostatic field of the others. The field is updated repeatedly until the input and output charge distributions agree. This was an important computational innovation, but the simple product wave function did not possess the required antisymmetry under exchange of identical electrons. The missing antisymmetry meant that exchange effects were not properly included.

John C. Slater introduced the determinantal representation of antisymmetric many-electron wave functions in 1929. In 1930, Vladimir Fock derived self-consistent-field equations that included exchange; Slater independently developed closely related ideas. The determinant changes sign when two electron coordinates are interchanged and vanishes if two occupied spin orbitals are identical. Thus, the Pauli principle is built into the mathematical form of the wave function rather than imposed as an additional occupation rule after the calculation. The resulting theory became known as the Hartree–Fock approximation.

A decisive development for molecules was the expression of molecular orbitals as linear combinations of known basis functions. In 1951, Clemens C. J. Roothaan and George G. Hall independently formulated matrix versions of the closed-shell equations. Their work connected Hartree–Fock theory with molecular orbital theory and made it suitable for general-purpose computer implementation. Instead of solving differential and integral equations directly in three-dimensional space, a program could optimize a finite set of orbital coefficients.

Subsequent advances made the method a standard component of computational chemistry: analytical integral evaluation with Gaussian functions, efficient matrix diagonalization, direct evaluation of electron-repulsion integrals, analytical derivatives, and improved convergence algorithms. Hartree–Fock also developed in parallel within atomic, solid-state, and nuclear physics. Although electron correlation and density-based approaches have displaced it for many routine quantitative predictions, its equations remain foundational for understanding exchange, independent-particle descriptions, and the construction of systematically improved many-body approximations.

The variational single-determinant approximation

For a molecule, the usual starting point is the Born–Oppenheimer approximation, which treats nuclear positions as fixed while the electronic problem is solved. In nonrelativistic atomic units, the electronic Hamiltonian operator is

H^e=∑i=1N[−12∇i2−∑AZAriA]+∑i<j1rij.\hat H_{\mathrm e} = \sum_{i=1}^{N} \left[ -\frac12\nabla_i^2 -\sum_A\frac{Z_A}{r_{iA}} \right] + \sum_{i<j}\frac{1}{r_{ij}}.

Here NN is the number of electrons, ZAZ_A is the charge number of nucleus AA, riAr_{iA} is an electron–nucleus distance, and rijr_{ij} is an electron–electron distance. The nuclear repulsion energy, ENN=∑A<BZAZB/RABE_{\mathrm{NN}}=\sum_{A<B}Z_AZ_B/R_{AB}, is added to obtain the total energy at that geometry. Atomic units set the electron mass, elementary charge magnitude, reduced Planck constant, and Coulomb constant to unity.

The Hartree–Fock trial state is

Φ(x1,…,xN)=1N!det⁡[χi(xj)],\Phi(x_1,\ldots,x_N) = \frac{1}{\sqrt{N!}} \det[\chi_i(x_j)],

where xx denotes both spatial and spin coordinates and the spin orbitals satisfy ⟨χi∣χj⟩=δij\langle\chi_i|\chi_j\rangle=\delta_{ij}. The determinant contains all permutations needed to antisymmetrize the product. Its form embodies the indistinguishability of electrons, which are fermions, and connects the spatial arrangement of electrons with their spin angular momentum. A determinant is not generally an eigenfunction of total spin squared, although commonly used restricted closed-shell determinants are.

The orbitals are chosen using the variational principle:

EHF=min⁡{χi}⟨Φ∣H^e∣Φ⟩,E_{\mathrm{HF}} = \min_{\{\chi_i\}} \langle\Phi|\hat H_{\mathrm e}|\Phi\rangle,

subject to orbital orthonormality and any imposed symmetry or spin restrictions. For the ground state of the same Hamiltonian, a normalized trial function gives an energy no lower than the exact ground-state energy. Consequently, the fully optimized Hartree–Fock energy is an upper bound. An actual self-consistent calculation, however, may find a local minimum or another stationary point rather than the global determinant minimum.

Optimization within a determinant is much more restrictive than optimization over all antisymmetric wave functions. The orbitals may adapt to the average charge distribution, but the state cannot describe arbitrary instantaneous correlations among electrons. A determinant is not an uncorrelated state in the everyday sense: antisymmetry already creates exchange correlations between electrons of the same spin. The conventional statement that Hartree–Fock “neglects correlation” refers to correlation beyond this determinant structure.

Unitary transformations among occupied orbitals change the determinant at most by an overall phase and therefore leave its density and energy unchanged. The physically essential object is the occupied orbital subspace, not one unique set of occupied orbitals. Canonical and localized occupied orbitals can describe exactly the same Hartree–Fock state. This freedom becomes important when interpreting chemical bonds, comparing orbital pictures, and constructing local correlation methods.

Coulomb interaction, exchange, and the Fock equations

The expectation value of the electronic Hamiltonian in a determinant is

EHF=∑ihii+12∑ij(Jij−Kij),E_{\mathrm{HF}} = \sum_i h_{ii} + \frac12\sum_{ij}(J_{ij}-K_{ij}),

where hii=⟨χi∣h^∣χi⟩h_{ii}=\langle\chi_i|\hat h|\chi_i\rangle, with h^\hat h the one-electron kinetic-plus-nuclear-attraction operator. The Coulomb and exchange integrals are

Jij=∬∣χi(x1)∣2∣χj(x2)∣2r12 dx1dx2,J_{ij} = \iint \frac{|\chi_i(x_1)|^2|\chi_j(x_2)|^2} {r_{12}}\,dx_1dx_2,
Kij=∬χi∗(x1)χj(x1)χj∗(x2)χi(x2)r12 dx1dx2.K_{ij} = \iint \frac{\chi_i^*(x_1)\chi_j(x_1) \chi_j^*(x_2)\chi_i(x_2)} {r_{12}}\,dx_1dx_2.

The Coulomb term resembles the classical electrostatic interaction of charge distributions. Exchange has no equivalent interpretation as an additional classical force; it arises from the antisymmetry of the wave function.

Variation of the orbitals under orthonormality constraints gives

f^χi=∑jλjiχj.\hat f\chi_i=\sum_j\lambda_{ji}\chi_j.

The Hermitian matrix of Lagrange multipliers can be diagonalized by an occupied-orbital rotation. In this canonical representation, the Fock operator satisfies

f^χi=εiχi,f^=h^+∑j(J^j−K^j).\hat f\chi_i=\varepsilon_i\chi_i, \qquad \hat f=\hat h+\sum_j(\hat J_j-\hat K_j).

These equations resemble one-electron eigenvalue equations, but the operator itself depends on the orbitals being sought. They are therefore nonlinear equations for the complete orbital set.

The Coulomb operator acts by multiplying a test function by the electrostatic potential of an occupied orbital. The exchange operator instead acts nonlocally:

(K^jχi)(x1)=χj(x1)∫χj∗(x2)χi(x2)r12 dx2.(\hat K_j\chi_i)(x_1) = \chi_j(x_1) \int \frac{\chi_j^*(x_2)\chi_i(x_2)} {r_{12}}\,dx_2.

Its value at one point depends on the test orbital over all space. For ordinary collinear spin orbitals, exchange integrals between opposite-spin electrons vanish after spin integration. Exchange therefore reduces the pair probability for same-spin electrons at short separations, producing an exchange hole. Opposite-spin electrons have no corresponding exchange hole, although their Coulomb repulsion still generates correlation in the exact state.

For i=ji=j, Jii=KiiJ_{ii}=K_{ii}. Thus, the spurious Coulomb interaction of an occupied spin orbital with itself cancels exactly in the Hartree–Fock energy and equations. This makes Hartree–Fock exact for a one-electron system in a complete basis for the selected one-electron Hamiltonian. The cancellation should not be confused with a complete treatment of electron interaction: each electron still lacks the full response to the instantaneous positions of the others.

The factor of one-half in the energy avoids counting each electron pair twice. Accordingly, the total electronic energy is not simply the sum of occupied orbital energies:

EHF=∑iεi−12∑ij(Jij−Kij).E_{\mathrm{HF}} = \sum_i\varepsilon_i - \frac12\sum_{ij}(J_{ij}-K_{ij}).

Each orbital eigenvalue already contains interactions with the occupied electrons. Adding those eigenvalues without the correction would double-count their net pair interaction.

Basis sets and the matrix formulation

For molecular calculations, each spatial orbital is usually expanded in an electronic-structure basis set:

ϕi(r)=∑μ=1MCμiφμ(r).\phi_i(\mathbf r) = \sum_{\mu=1}^{M}C_{\mu i}\varphi_\mu(\mathbf r).

The functions φμ\varphi_\mu are commonly centered on nuclei and resemble atomic orbitals, although numerical grids, plane waves, and other representations are also possible. Substitution into the orbital equations gives the Roothaan–Hall equations for a restricted closed-shell system:

FC=SCε,\mathbf F\mathbf C = \mathbf S\mathbf C\boldsymbol{\varepsilon},

where S\mathbf S is the overlap matrix and F\mathbf F is the Fock matrix. Because atom-centered basis functions are generally nonorthogonal, this is a generalized matrix eigenvalue problem.

For real basis functions and doubly occupied spatial orbitals, a common density-matrix convention is

Pμν=2∑i∈occCμiCνi.P_{\mu\nu} = 2\sum_{i\in\mathrm{occ}}C_{\mu i}C_{\nu i}.

The Fock matrix is then

Fμν=hμν+∑λσPλσ[(μν∣λσ)−12(μλ∣νσ)],F_{\mu\nu} = h_{\mu\nu} + \sum_{\lambda\sigma}P_{\lambda\sigma} \left[ (\mu\nu|\lambda\sigma) -\frac12(\mu\lambda|\nu\sigma) \right],

with electron-repulsion integrals defined by

(μν∣λσ)=∬φμ(r1)φν(r1)φλ(r2)φσ(r2)r12 dr1dr2.(\mu\nu|\lambda\sigma) = \iint \frac{ \varphi_\mu(\mathbf r_1)\varphi_\nu(\mathbf r_1) \varphi_\lambda(\mathbf r_2)\varphi_\sigma(\mathbf r_2) }{r_{12}}\, d\mathbf r_1d\mathbf r_2.

In this convention, the electronic energy is Ee=12∑μνPμν(hμν+Fμν)E_{\mathrm e}=\tfrac12\sum_{\mu\nu}P_{\mu\nu}(h_{\mu\nu}+F_{\mu\nu}). Other conventions absorb the factor of two into the density or use spin-resolved matrices.

Slater-type functions have exponential radial decay and can reproduce the electron–nucleus cusp characteristic of exact Coulombic wave functions. Gaussian-type functions lack both the correct cusp and the correct long-range exponential form, but products of Gaussians have properties that simplify multicenter integrals enormously. Contracted Gaussian basis functions combine several primitive Gaussians to approximate useful atomic shapes. This computational advantage explains their widespread use in molecular programs.

Basis quality controls a separate error from the Hartree–Fock approximation itself. Minimal bases provide little flexibility; split-valence and multiple-zeta bases allow radial adjustment; polarization functions add higher angular momentum; and diffuse functions describe spatially extended density, especially in anions and weakly bound states. For nested variational spaces and properly optimized solutions, increasing the basis cannot raise the minimum energy. The complete-basis limit of Hartree–Fock nevertheless remains above the exact nonrelativistic energy for a correlated many-electron system.

Very diffuse or redundant functions can make the overlap matrix nearly singular, requiring careful numerical treatment. Finite atom-centered bases also produce basis-set superposition error: fragments in a complex can borrow basis functions from one another and gain an artificial energetic advantage relative to isolated-fragment calculations. Basis enlargement and counterpoise calculations address this representation error, not the missing physical correlation. A calculation can therefore be nearly converged with respect to its basis while still failing to describe an important interaction.

Restricted, unrestricted, and open-shell formulations

Restricted Hartree–Fock, usually abbreviated RHF, assigns paired electrons of opposite spin to the same spatial orbital. For an even-electron closed-shell system, each occupied spatial orbital contains one alpha-spin and one beta-spin electron. This restriction produces a spin-pure singlet determinant and reduces the computational work. It is often suitable near the equilibrium geometries of closed-shell molecules whose electronic structures are dominated by a single configuration. Pairing is nevertheless an imposed constraint, not a generally exact property of an interacting-electron state.

Unrestricted Hartree–Fock, or UHF, optimizes alpha and beta spatial orbitals independently. Its coupled equations can be written schematically as

f^α=h^+J^[ρα+ρβ]−K^α,f^β=h^+J^[ρα+ρβ]−K^β.\hat f^\alpha = \hat h+\hat J[\rho_\alpha+\rho_\beta]-\hat K^\alpha, \qquad \hat f^\beta = \hat h+\hat J[\rho_\alpha+\rho_\beta]-\hat K^\beta.

Both spin populations experience the Coulomb potential of the total electron density, whereas exchange acts within each spin population. UHF is widely used for radicals and open-shell atoms and can also describe spatially separated opposite-spin electrons by allowing their orbitals to differ.

The additional freedom means that the optimized UHF energy cannot exceed the corresponding restricted minimum when both use compatible basis and occupation constraints. However, a UHF determinant with fixed numbers of alpha and beta electrons is generally an eigenfunction of S^z\hat S_z, but not of S^2\hat S^2. Its expectation value ⟨S^2⟩\langle\hat S^2\rangle can exceed the value S(S+1)S(S+1) associated with the intended pure-spin state. This effect, called spin contamination, reflects admixture of other total-spin components and can complicate energies, spin densities, and subsequent correlated calculations.

Restricted open-shell Hartree–Fock, or ROHF, retains common spatial orbitals for paired electrons and introduces singly occupied orbitals for unpaired electrons. Standard high-spin ROHF constructions can preserve the desired total-spin character while avoiding the independent alpha and beta orbital sets of UHF. The stationary energy and occupied subspaces are well defined, but different choices of effective canonicalization can yield different orbital energies. Consequently, the detailed ROHF orbital-energy convention matters when those energies enter a subsequent approximation.

Generalized Hartree–Fock permits still more flexible spinors, including mixtures of alpha and beta character and noncollinear spin arrangements. Spatial symmetry may likewise be imposed or relaxed. Relaxing a constraint can lower the variational energy by allowing a determinant to imitate some effects of a more complicated wave function, but it may sacrifice a symmetry possessed by the exact finite-system state. The choice among restricted, unrestricted, and generalized forms therefore concerns the allowed variational space, not merely numerical convenience.

Self-consistent solution and computational cost

A Hartree–Fock calculation normally begins with the nuclear geometry, electron count, spin specification, basis functions, and an initial density matrix. Initial guesses may come from the one-electron core Hamiltonian, superposed atomic densities, fragment calculations, or a nearby converged geometry. The program constructs a Fock matrix from that density, solves the generalized eigenvalue equations, selects occupied orbitals, and forms an updated density. The cycle continues until selected measures of self-consistency fall below numerical thresholds.

Energy convergence alone is insufficient to establish that the orbitals satisfy the equations accurately. Common tests include changes in the density matrix and the residual

R=FPS−SPF.\mathbf R = \mathbf F\mathbf P\mathbf S - \mathbf S\mathbf P\mathbf F.

For a stationary closed-shell solution, this residual vanishes. Small energy changes can occur even when the orbital gradient remains appreciable, particularly in difficult near-degenerate cases. Calculations of derivatives and response properties often require tighter convergence than calculations concerned only with a rough total energy.

Simple replacement of the old density by the newly generated density can oscillate or diverge. Damping mixes successive densities, while level shifting modifies virtual-orbital energies during iteration to discourage unstable changes of occupation or orbital character. Peter Pulay introduced direct inversion in the iterative subspace, commonly called DIIS, in 1980. DIIS combines information from previous iterations to reduce an extrapolated residual and is a standard acceleration technique. More robust alternatives include energy-based extrapolation, augmented-Hessian methods, and second-order orbital optimization.

Convergence does not establish uniqueness or stability. Different starting guesses may lead to different charge distributions, spin arrangements, or orbital occupations. A Hartree–Fock stability analysis examines the second derivative of the energy with respect to occupied–virtual orbital rotations. Negative curvature identifies a direction in which the energy can decrease. A solution stable within RHF can still be unstable to a spin-symmetry-breaking UHF perturbation. Excited determinants and metastable electronic configurations may also be intentionally targeted, but these require occupation control rather than automatic filling of the lowest canonical levels.

With MM basis functions, the number of four-index electron-repulsion integrals scales formally as M4M^4, and conventional Fock construction has corresponding quartic work. Dense diagonalization scales approximately as M3M^3. Storing all integrals becomes impractical for large bases, so direct SCF algorithms recompute integrals as needed and use bounds to screen negligible contributions. Density fitting, Cholesky decompositions, localized representations, and sparse matrix techniques reduce storage or computational effort. Their practical scaling depends on dimensionality, basis locality, energy gaps, and screening thresholds; a nominal scaling law alone does not predict the cost of every system.

Energies, orbitals, and measurable properties

The Hartree–Fock density provides expectation values of one-electron operators. With the spin-summed density matrix defined above,

ρ(r)=∑μνPμνφμ(r)φν(r).\rho(\mathbf r) = \sum_{\mu\nu}P_{\mu\nu} \varphi_\mu(\mathbf r)\varphi_\nu(\mathbf r).

Integration gives the number of electrons. Combining this density with the nuclear charges yields quantities such as the electric dipole moment. Similar procedures give quadrupole moments and other static properties. Atomic charges obtained by partitioning the density are less direct: different population-analysis schemes divide the same density differently, so a particular assigned charge is not a uniquely defined observable.

Canonical orbital energies are Lagrange multipliers associated with orbital orthonormality. They are useful descriptors, but they are not generally electron addition, electron removal, or optical excitation energies. Koopmans’ theorem supplies a limited connection: for removal of an electron from occupied orbital ii, the frozen-orbital Hartree–Fock ionization energy is

Iifrozen=−εi.I_i^{\mathrm{frozen}}=-\varepsilon_i.

The theorem assumes that the remaining orbitals do not relax and compares determinantal energies without adding correlation corrections. It therefore provides an approximation to ionization energy, rather than an exact identification of an orbital eigenvalue with an experimental measurement.

Orbital relaxation generally lowers the energy of the ion relative to a frozen-orbital description, while correlation changes the neutral and ionic energies differently. Their contributions can partly cancel, explaining why some occupied orbital energies are useful estimates. Virtual orbital energies are particularly problematic as electron affinities: the unoccupied orbitals are generated in the field of the original occupied state, not optimized for an added electron. A difference of separately optimized total energies, often called a delta-SCF calculation, includes orbital relaxation but remains subject to the underlying approximation and possible differences in state character.

Molecular structures and forces follow from derivatives of the total energy with respect to nuclear coordinates. At an optimized electronic solution, variational stationarity simplifies these derivatives, but atom-centered basis functions move when nuclei move. Their derivatives generate Pulay terms, which must be included alongside the explicit derivative of the Hamiltonian. Analytical gradients enable geometry optimization and force calculations; second derivatives give harmonic vibrational frequencies. These predictions omit correlation and, unless separately included, effects such as anharmonicity, nuclear quantum motion beyond the harmonic model, and relativity.

Properties describing a response to an applied perturbation require the orbitals themselves to respond. Polarizabilities, for example, are not obtained accurately merely by holding the ground-state density fixed in an external electric field. Coupled-perturbed Hartree–Fock equations determine the first-order orbital response. Related linear-response equations underlie time-dependent Hartree–Fock excitation calculations. They include collective rearrangement of the determinant, but their accuracy remains limited by the single-determinant ground-state approximation and the restricted form of the excitation treatment.

Correlation, bond dissociation, and characteristic failures

For a specified nonrelativistic Hamiltonian and a consistently defined Hartree–Fock reference, the correlation energy is conventionally

Ecorr=Eexact−EHF.E_{\mathrm{corr}} = E_{\mathrm{exact}}-E_{\mathrm{HF}}.

It is nonpositive when EHFE_{\mathrm{HF}} is the corresponding variational minimum. Comparisons must hold the Hamiltonian, basis treatment, and reference convention fixed: changing relativistic assumptions or comparing a finite-basis result with a complete-basis benchmark mixes distinct errors. Correlation energy is usually a modest fraction of the total electronic energy, but chemical reaction energies and intermolecular binding energies are small differences between large totals and can depend critically on it.

Dynamic correlation describes the rapid avoidance of electrons beyond that enforced by exchange. The exact wave function responds to the instantaneous separation of electrons, including opposite-spin pairs; a determinant cannot generally represent this dependence. For a Coulomb interaction, the exact wave function has a characteristic electron–electron cusp at coalescence. A single product structure lacks the required general pair dependence, even when its one-electron orbitals are represented in a very large basis. Basis completeness alone therefore cannot eliminate correlation error.

Static or nondynamic correlation becomes important when several configurations have comparable weights. The stretched hydrogen molecule is the standard example. Near equilibrium, a restricted determinant with the bonding orbital doubly occupied gives a useful description. At large separation, however, that determinant retains unphysical ionic contributions in addition to the desired state with one electron on each atom. A proper spin-singlet dissociation state requires more than one determinant. Increasing the RHF basis does not remove this structural limitation.

Beyond a certain separation, UHF can lower the energy by localizing opposite-spin electrons predominantly on different atoms. The separation at which the restricted solution develops the relevant unrestricted instability is called a Coulson–Fischer point. For separated hydrogen atoms, this symmetry-broken approach reaches the correct energy limit, but its determinant is not the pure molecular singlet. The example illustrates a recurring distinction: an improved energy can result from breaking symmetry without producing a correspondingly faithful many-electron state. Transition-metal complexes, diradicals, and bond-breaking reactions often exhibit more complicated versions of the same issue.

Long-range London dispersion provides another characteristic failure. For well-separated neutral, nonpolar, approximately spherical fragments, the leading attractive contribution has the form −C6/R6-C_6/R^6, arising from correlated charge fluctuations. Ordinary Hartree–Fock interaction energies lack this term. They can include permanent electrostatics, induction, and exchange effects, but not general dispersion attraction. Consequently, descriptions of van der Waals forces, molecular packing, and weakly bound complexes often require correlation methods or additional dispersion treatments. Correlation error also commonly affects bond lengths, reaction barriers, vibrational frequencies, and electron affinities, although its magnitude and direction are system dependent.

Hartree–Fock as a reference for correlated theories

Many post-Hartree–Fock methods retain Hartree–Fock orbitals as a convenient one-particle basis while expanding the many-electron wave function beyond one determinant. Occupied orbitals define the reference configuration, and virtual orbitals provide destinations for electronic excitations. A determinant formed by replacing one occupied orbital is called a single excitation; replacing two produces a double excitation. These are algebraic components of a wave-function expansion and do not necessarily represent independently observable excitation processes.

A key consequence of orbital optimization is Brillouin’s theorem:

⟨Φia∣H^∣Φ0⟩=0,\langle\Phi_i^a|\hat H|\Phi_0\rangle=0,

for singly excited determinants generated by allowed occupied–virtual rotations of a stationary Hartree–Fock reference. Thus, the reference does not couple directly through the Hamiltonian to those single excitations. This does not make single excitations universally irrelevant: they can participate indirectly in correlated wave functions, represent relaxation relative to another reference, and become essential in excited-state descriptions. The theorem expresses stationarity of the determinant, not completeness of its electronic description.

Møller–Plesset perturbation theory partitions the Hamiltonian using the Hartree–Fock Fock operator and treats the residual interaction through perturbation theory. Its second-order form, MP2, is widely used for dynamic correlation and includes long-range dispersion absent from Hartree–Fock. For canonical spin orbitals,

EMP2(2)=14∑ijocc∑abvirt∣⟨ij∥ab⟩∣2εi+εj−εa−εb.E_{\mathrm{MP2}}^{(2)} = \frac14 \sum_{ij}^{\mathrm{occ}} \sum_{ab}^{\mathrm{virt}} \frac{|\langle ij\Vert ab\rangle|^2} {\varepsilon_i+\varepsilon_j-\varepsilon_a-\varepsilon_b}.

The antisymmetrized integrals in the numerator couple occupied pairs to virtual pairs. Small energy denominators signal possible failure of the perturbative treatment, particularly near degeneracy or bond dissociation.

Configuration interaction constructs and diagonalizes the Hamiltonian in a selected space of determinants or spin-adapted configurations. Full configuration interaction gives the exact solution within the chosen finite one-electron basis, but its cost grows combinatorially. Truncated configuration interaction can be variational, yet common truncations fail to give additive energies for separated noninteracting fragments. Coupled-cluster theory instead uses an exponential ansatz, ∣Ψ⟩=eT^∣Φ0⟩|\Psi\rangle=e^{\hat T}|\Phi_0\rangle, where excitation operators in T^\hat T generate connected correlation contributions and their products. Standard coupled-cluster energies are not generally variational, but their size-extensive structure is a major advantage.

A single Hartree–Fock reference is most effective when one determinant dominates the state. Where several configurations are essential, multiconfigurational self-consistent-field methods optimize both orbitals and configuration coefficients. Complete-active-space self-consistent field treats all configurations within a selected active orbital space, after which additional methods can recover external dynamic correlation. The quality of the reference affects perturbative denominators, convergence, and the interpretation of excitation amplitudes. Hartree–Fock therefore serves not only as an approximate endpoint, but also as a diagnostic and organizational framework for choosing a more appropriate many-electron description.

Density-functional, periodic, relativistic, and time-dependent forms

Hartree–Fock and density functional theory often use similar computational machinery: both can involve orbitals, density matrices, basis expansions, and self-consistent iteration. Their formal meanings differ. Hartree–Fock minimizes the expectation value of the interacting Hamiltonian over determinants. Kohn–Sham density-functional theory introduces an auxiliary noninteracting system that reproduces the interacting ground-state density in the exact theory, with exchange and correlation incorporated through a functional. The Kohn–Sham determinant is therefore not generally intended as an approximation to the full interacting wave function in the same sense as a Hartree–Fock determinant.

Hybrid density functionals incorporate a fraction of Hartree–Fock-like nonlocal exchange together with density-functional exchange and correlation. Range-separated hybrids divide the electron interaction into short- and long-range parts and apply different treatments to them. In this context, “exact exchange” means the exact exchange energy of the selected determinant or orbital system; it does not imply exact total energy or inclusion of exact correlation. Mixing exchange contributions changes the equations and their predictions, rather than simply adding a universal correction to Hartree–Fock.

For periodic solids, orbitals are represented by Bloch functions and the exchange operator includes contributions from occupied states across reciprocal space. Hartree–Fock can describe band formation and magnetic ordering, but the long range of Coulomb exchange makes periodic calculations demanding. Brillouin-zone sampling and singularity treatment require care. Because correlation screens electron interactions in real materials, unscreened Hartree–Fock often overestimates insulating gaps and gives problematic behavior for metals. Its canonical orbital spectrum should not be identified directly with an exact quasiparticle band structure.

Relativistic self-consistent-field theories replace the nonrelativistic one-electron operator with a relativistic description. Dirac–Hartree–Fock uses four-component spinors and incorporates spin–orbit coupling naturally, while two-component approximations seek lower computational cost. These methods are especially important for heavy elements, where relativistic changes in orbital energies and radial distributions strongly affect bonding and spectra. Their practical formulation must separate the electronic sector appropriately from negative-energy states; simply minimizing over unrestricted four-component functions would not be a valid extension of the nonrelativistic variational problem.

Time-dependent Hartree–Fock allows occupied orbitals to evolve under a Fock operator constructed from the instantaneous density. In an orthonormal one-particle representation, the density matrix obeys

idPdt=[F[P],P]i\frac{d\mathbf P}{dt} = [\mathbf F[\mathbf P],\mathbf P]

in atomic units. Linearizing around a stationary state gives the time-dependent Hartree–Fock response equations, closely associated with the random-phase approximation including exchange. This framework describes collective electronic response and certain excitation spectra, while real-time propagation treats driven dynamics within the determinant manifold. Processes requiring strong dynamical correlation, important multiple-excitation character, or substantial branching into distinct electronic configurations lie beyond that manifold.