Exchange-Correlation Functional

The simplest rung, where the XC energy depends only on the local electron density at each point in space. Assumes the system behaves locally like a uniform electron gas.

• Ingredients: ρ(r) only
• Strengths: Exact for uniform electron gas; surprisingly accurate for simple metals and geometries with slowly varying density
• Weaknesses: Overbinds molecules; fails for hydrogen bonds and systems with strong density inhomogeneity
• Example: VWN, PW92 correlation functionals
• Historical significance: The original functional that proved DFT could work in practice

Introduces the gradient of the electron density (∇ρ) as an additional ingredient, capturing the inhomogeneity of real molecular systems.

• Ingredients: ρ(r) and ∇ρ(r)
• Key functionals: PBE (Perdew-Burke-Ernzerhof), BLYP, PW91
• PBE: The workhorse of solid-state physics; parameter-free by construction
• BLYP: Combines Becke88 exchange with LYP correlation; historically dominant in quantum chemistry
• Improvement over LDA: Corrects overbinding; better hydrogen bond energies; reasonable thermochemistry
• Remaining error: Self-interaction error persists; van der Waals interactions absent

Adds the kinetic energy density (τ) and/or the Laplacian of the density (∇²ρ) to the GGA ingredients, enabling detection of covalent bonds, metallic bonds, and weak interactions.

• Ingredients: ρ(r), ∇ρ(r), and τ(r) or ∇²ρ(r)
• Key functionals: TPSS, SCAN, M06-L
• SCAN (Strongly Constrained and Appropriately Normed): Satisfies all 17 known exact constraints for meta-GGAs; accurately describes covalent, metallic, ionic, and hydrogen bonds without empiricism
• Advantage: Better description of intermediate-range van der Waals interactions and diverse bonding types
• Computational cost: Marginally higher than GGA; negligible overhead for most systems

Incorporates a fraction of exact Hartree-Fock exchange into the XC functional, partially eliminating the self-interaction error that plagues lower rungs.

• Ingredients: ρ(r), ∇ρ(r), τ(r), and occupied orbitals ψᵢ(r)
• Key functionals: B3LYP, PBE0, HSE06
• B3LYP: The most cited functional in history; mixes Becke88 exchange, exact HF exchange, and LYP correlation with three empirical parameters
• PBE0: A non-empirical hybrid with 25% exact exchange; excellent for band gaps and reaction barriers
• HSE06: A screened hybrid that applies exact exchange only at short range, dramatically reducing cost for periodic solids
• Cost: Significantly more expensive than GGAs due to exact exchange evaluation; scales poorly with system size

The highest rung, incorporating unoccupied orbitals via second-order perturbation theory (MP2) or the Random Phase Approximation (RPA) to capture long-range correlation effects.

• Ingredients: ρ(r), ∇ρ(r), τ(r), occupied ψᵢ(r), and unoccupied ψₐ(r) orbitals
• Key functionals: B2PLYP, DSD-PBEP86, ωB97X-2
• Double hybrids: Mix exact HF exchange with MP2-like correlation; achieve near-chemical accuracy (~1 kcal/mol) for thermochemistry
• RPA: A fully non-local correlation treatment that naturally captures van der Waals interactions without empirical dispersion corrections
• Cost: Most expensive rung; MP2 step scales as O(N⁵) formally; practical for small to medium molecules
• Status: The current frontier of practical DFT accuracy before wavefunction methods

The conceptual limit of the ladder is the exact XC functional, which would reproduce the full solution of the many-body Schrödinger equation at DFT cost. This remains unknown and is the holy grail of DFT development.

• Machine learning functionals: Neural networks trained on exact XC potentials from small systems (e.g., DM21, DeepMind's functional) represent a new paradigm that bypasses the ladder entirely
• Constraints-based development: Each rung satisfies more exact physical constraints; the exact functional would satisfy all of them
• Current frontier: ML-trained functionals that learn the map from density to XC energy directly from high-level wavefunction data
• Philosophical shift: The ladder may be superseded by data-driven functional discovery rather than analytic construction

The practical foundation of modern DFT. These single-particle equations map the complex, interacting many-body problem onto a fictitious system of non-interacting electrons that yields the same ground-state density. The only unknown term in this elegant reformulation is the exchange-correlation functional, making it the sole point of approximation. Solving these equations self-consistently is the core computational loop of any DFT code.

The iterative computational procedure that solves the Kohn-Sham equations. The process begins with an initial guess for the electron density, constructs the effective potential (including the exchange-correlation contribution), solves for new orbitals, and computes a new density. This cycle repeats until the input and output densities converge within a tight tolerance. The choice of functional directly impacts the number of SCF cycles required and the stability of convergence.

A hierarchical classification of exchange-correlation functionals proposed by John Perdew, organized by increasing complexity and theoretical sophistication. Each rung introduces more physical ingredients from the density:

• LDA: Local density only
• GGA: Adds density gradient
• meta-GGA: Adds kinetic energy density
• Hybrid: Mixes in exact Hartree-Fock exchange
• Double-Hybrid: Adds unoccupied orbitals (MP2 correlation) Climbing the ladder generally improves accuracy but increases computational cost.

A set of mathematical functions used to represent the Kohn-Sham molecular orbitals. The exchange-correlation functional is evaluated on a numerical grid, but the orbitals themselves are expanded in this basis. Common types include:

• Localized Gaussian-type orbitals (GTOs): Dominant in molecular codes
• Plane waves: Natural for periodic solids
• Numerical atomic orbitals: Used for efficiency in large systems The choice of basis set and functional are deeply intertwined; a functional parameterized with one basis may perform poorly with another.

Often called the 'gold standard' of quantum chemistry. This highly accurate wavefunction-based method systematically accounts for electron correlation through an exponential excitation operator. It is not a DFT method, but it serves as the critical benchmark for generating high-fidelity reference data used to train machine learning potentials and to parameterize or validate new exchange-correlation functionals. Its extreme computational cost (scaling as O(N⁷)) limits it to small systems.

An extension of ground-state DFT for calculating excited-state properties and electronic spectra. TD-DFT requires a time-dependent exchange-correlation functional, known as the exchange-correlation kernel, which describes the dynamic response of the electron density to a time-varying external field. The accuracy of predicted excitation energies and oscillator strengths is critically dependent on the quality of this kernel, with standard adiabatic approximations often failing for charge-transfer states.