Configuration Interaction: The Basic Idea

Introduction

Hartree–Fock selects one optimised determinant. When electrons move in correlated ways or several configurations are close in energy, one determinant may be insufficient. Configuration interaction, CI, keeps the Pauli-respecting determinant building blocks but adds them together with coefficients determined by a variational eigenvalue problem. The method is conceptually direct: a larger wavefunction space can describe more electronic patterns, though its size grows rapidly.

Core explanation

Write an electronic wavefunction as Ψ CI=Σ I c IΦ I, where Φ I are Slater determinants or spin-adapted configuration state functions built from a selected one-electron orbital basis. Each determinant is antisymmetric, so their linear combination is also antisymmetric. Coefficients c I are found by diagonalising the many-electron Hamiltonian matrix H IJ=⟨Φ I Ĥ Φ J⟩, with an overlap matrix if the configurations are not orthonormal. Different eigenvectors approximate different electronic states within the chosen configuration space.

Starting from a Hartree–Fock reference determinant, other configurations can be described as single, double, triple and higher excitations, according to how many occupied spin orbitals are replaced by virtual ones. The word excitation here labels a determinant relative to a reference; it does not mean the final CI state must be an electronically excited molecule. Double and higher substitutions can contribute to correlation of a ground state. Spin and spatial symmetry restrict which configurations can mix with the target state.

Full CI includes every allowed determinant in a fixed finite one-electron basis for the chosen electron count and symmetry. It gives the exact solution of the electronic Schrödinger equation within that finite basis and chosen Hamiltonian, aside from numerical error. It is not the exact physical molecule in an infinite basis, and it does not automatically include relativistic, solvent or nuclear-motion effects. The number of determinants grows combinatorially, making full CI practical only for small systems or limited active spaces.

Truncated CI includes only selected excitation ranks or configurations, such as singles and doubles relative to a reference. The variational principle means adding more configurations to a nested CI space cannot raise its lowest energy. A truncated calculation can therefore improve on the reference energy, but it may miss important higher excitations or near-degenerate determinants not chosen. Standard truncated CI also has a size-extensivity problem: the energy of two noninteracting copies need not equal twice the separately computed energy, which limits some large-system energy comparisons.

The simplest two-configuration model shows the energy-lowering mechanism. If determinants have diagonal energies E₁ and E₂ and Hamiltonian coupling V, the eigenvalues of [[E₁,V],[V,E₂]] are (E₁+E₂)/2 ± √{[(E₁−E₂)/2]²+V²}. The lower eigenvalue is no higher than the lower uncoupled diagonal energy and decreases when nonzero coupling mixes the states. Symmetry can force V=0, in which case the determinants do not mix through that Hamiltonian element.

CI coefficients depend on the chosen orbital and configuration basis. A coefficient near one for a reference determinant suggests a single-reference picture may be useful, but coefficients alone can be basis dependent. Several large coefficients often signal important static correlation. Observables and energy differences should be checked for convergence with both determinant space and orbital basis.

CI is especially instructive near bond dissociation, where ionic and covalent arrangements may both need representation. A spin-adapted combination of configurations can preserve the intended total spin while describing electrons localising on separate fragments. The method also supports excited-state calculations, but state ordering and transition intensities require their own convergence and symmetry checks.

Step-by-step reasoning

Choose a one-electron orbital basis and an antisymmetric reference. Generate configurations with appropriate electron count and target spin/spatial symmetry. Build the Hamiltonian matrix, diagonalise it and inspect eigenvalues and coefficients. Increase the configuration space and basis to check convergence, then remember that a finite-basis full CI result is exact only within its stated Hamiltonian and basis.

Visual explanation

Draw one HF determinant as a central box, then arrows to boxes representing single and double orbital substitutions. Connect boxes with nonzero Hamiltonian matrix elements and show their weighted sum forming the CI state. A small two-by-two matrix beside them yields two mixed energy levels, with the lower one below the original reference diagonal energy.

Real-world analogy

A song played by one instrument may miss a complex harmony, while a weighted ensemble of instruments can reproduce it more faithfully. CI combines multiple valid determinant patterns into one state. The analogy captures expansion flexibility but not the variational energy equation or the need for antisymmetric electron wavefunctions.

Real-world example

As H₂ stretches, configurations with electrons distributed differently between bonding and antibonding orbitals become important for a correct singlet dissociation picture. A CI expansion can combine them rather than forcing one closed-shell determinant to carry the whole state. The required active configurations become more apparent when orbital gaps narrow.

Why?

Why does mixing generally lower the ground-state energy? The variational search over linear combinations includes each individual determinant as a special case. Allowing a nonzero coupling permits an optimised combination with energy at or below the best included single determinant. Symmetry-incompatible determinants can have zero coupling and may not contribute to that state.

Common misconception

Full CI in a small basis is not an exact solution of all physical chemistry; basis incompleteness and Hamiltonian approximations remain. Also, a determinant called a double excitation can help describe a ground state. Excitation rank is relative to a reference occupancy, not a claim about the observed spectral state.

Worked example

Take a two-configuration Hamiltonian with E₁=0, E₂=0.20 and V=0.10 energy units. Its lower eigenvalue is 0.10−√(0.10²+0.10²)≈−0.0414 units, below E₁=0. The upper is about 0.2414. Mixing has lowered the lowest variational energy, while the size of the effect depends on coupling and the energy separation of the configurations.

Quick check

1. Is full CI exact in an infinite physical sense? Answer: No. It is exact within a specified finite orbital basis and Hamiltonian, but basis and physical-model limits remain. 2. Can a double excitation contribute to a correlated ground-state CI wavefunction? Answer: Yes. The term double describes replacement of two occupied spin orbitals relative to a reference, not the energy rank of the resulting state.

Exam focus

Write Ψ as a determinant expansion and distinguish truncated from full CI. State the finite-basis qualification of exactness and the combinatorial cost. Use variational reasoning to explain ground-energy lowering, while noting symmetry constraints and possible size-extensivity errors in truncated CI.

Advanced insight

CI gives a linear variational ansatz, while coupled-cluster theory uses an exponential of excitation operators acting on a reference. The latter can restore size-extensive behaviour for common truncations but is not generally variational. Comparing methods requires attention to the property, reference quality and computational cost rather than ranking them by one slogan.

Summary

Configuration interaction mixes multiple antisymmetric determinants to describe electron correlation and near-degenerate states. Diagonalising their Hamiltonian matrix gives variational energies and state coefficients. Full CI is exact within a finite basis but grows combinatorially; truncated CI is more affordable and can miss configurations or suffer size-extensivity limitations.

Practice questions

1. Why is a linear combination of Slater determinants still antisymmetric under exchange of identical electrons? Answer: Every determinant changes sign on exchange, so their weighted sum also changes sign as a whole. 2. If two configurations have different exact spatial symmetries, can their Hamiltonian coupling be nonzero at a symmetry-preserving geometry? Answer: Not when their product lacks the totally symmetric species. Symmetry then forces the coupling matrix element to zero. 3. Why might a CI calculation with only singles and doubles fail at a strongly stretched bond? Answer: The reference may be qualitatively poor and additional near-degenerate or higher-rank configurations may matter; a limited truncation may not represent the required multi-reference state.