Density functional theory

Result

Density functional theory

T-111Home CU-401Threads quantum

Statement

Energy as a functional of the electron density.

Why it matters

hartree-fock already showed how to approximate a many-electron wavefunction by a self-consistent, mean-field treatment, but the Hartree–Fock wavefunction itself — a function of every electron's three spatial coordinates simultaneously — is an object of forbidding computational cost for anything beyond a fairly small molecule, and it still misses electron correlation beyond exchange entirely. Density functional theory offers a strikingly different, and for most practical chemistry far cheaper, route to the same goal: rather than solving for the full multi-electron wavefunction, it works with the electron density, a function of only three spatial coordinates regardless of how many electrons the system contains, while still capturing electron correlation, at least approximately, that Hartree–Fock leaves out.

This shift from wavefunction to density is what has made DFT, more than any other single method, the practical workhorse of modern computational chemistry, routinely used alongside potential-energy-surfaces calculations to locate molecular geometries, transition states, and reaction energetics for systems well beyond the reach of high-level wavefunction methods.

Hypotheses

The ground-state electron density \(\rho(\mathbf{r})\) uniquely determines the external potential (and hence, in principle, every ground-state property) of a many-electron system, up to a trivial additive constant.This is the content of the first Hohenberg–Kohn theorem, and it is what justifies working with the density at all: without this uniqueness guarantee, a given density could in principle correspond to more than one distinct physical system, and computing properties from density alone would be ill-defined. The exact ground-state energy functional \(E[\rho]\) is variational: it is minimised by, and only by, the true ground-state density.This is the second Hohenberg–Kohn theorem, and it supplies the practical route to actually finding the ground-state density: search over trial densities and minimise the resulting energy, exactly analogous in spirit to the variational principle already used for wavefunction-based methods like Hartree–Fock. The exact exchange-correlation functional, the one piece of \(E[\rho]\) that captures all of the genuinely many-body quantum effects not accounted for by the classical electrostatic and single-particle kinetic terms, is not known in closed form and cannot be derived exactly for a general system; every practical DFT calculation therefore relies on an approximate exchange-correlation functional, and the choice of functional (Corollaries) is the single largest source of variability in DFT results across different studies of the same system.

Proof

E[\rho] = T_s[\rho] + J[\rho] + E_{xc}[\rho] + \int v_{\text{ext}}(\mathbf{r})\rho(\mathbf{r})\,d\mathbf{r}

The Kohn–Sham approach partitions the total energy functional into a fictitious non-interacting kinetic energy \(T_s\), the classical Coulomb (Hartree) electron–electron repulsion \(J\), an exchange-correlation term \(E_{xc}\) collecting everything else quantum-mechanically subtle, and the interaction of the density with the external (typically nuclear) potential; this partition is exact in principle, with all the approximation concentrated entirely in the unknown \(E_{xc}\) term. B

\left[-\tfrac{1}{2}\nabla^2 + v_{\text{eff}}(\mathbf{r})\right]\psi_i(\mathbf{r}) = \varepsilon_i\,\psi_i(\mathbf{r}), \qquad \rho(\mathbf{r})=\sum_i|\psi_i(\mathbf{r})|^2

The Kohn–Sham equations replace the intractable many-body interacting problem with a fictitious system of non-interacting electrons moving in an effective potential \(v_{\text{eff}}\) (built from the nuclear potential, the classical Coulomb term, and the exchange-correlation potential \(v_{xc}=\delta E_{xc}/\delta\rho\)) chosen so this fictitious system's density exactly reproduces the true, interacting system's density; this is mathematically analogous to a one-electron Schrödinger equation and is solved self-consistently, since \(v_{\text{eff}}\) itself depends on the density it is used to compute. B

v_{\text{eff}}[\rho] \to \rho(\mathbf{r}) \to v_{\text{eff}}[\rho] \to \dots \quad\text{iterated to self-consistency}

Because the effective potential depends on the very density the Kohn–Sham orbitals are used to construct, the equations are solved iteratively: guess a starting density, build \(v_{\text{eff}}\), solve for new orbitals and a new density, rebuild \(v_{\text{eff}}\), and repeat until the density (and hence the energy) stops changing between iterations — the identical self-consistent-field logic already familiar from hartree-fock, applied here to a density rather than a wavefunction. A

E_{xc}[\rho] \approx E_{xc}^{\text{approx}}[\rho] \quad\text{(LDA, GGA, hybrid, etc.)}

Since the true \(E_{xc}\) is unknown, practical calculations substitute one of a hierarchy of increasingly sophisticated approximations: the local density approximation (LDA, depending only on the local density value), generalised gradient approximations (GGA, also including the local density gradient), and hybrid functionals (mixing in a fraction of exact Hartree–Fock exchange, e.g. B3LYP), each trading computational cost against accuracy differently, and each performing better or worse depending on the specific chemical system and property being computed. B

Result

\rho(\mathbf{r}) \ \longleftrightarrow\ \text{all ground-state properties}, \qquad E[\rho]\text{ minimised at the true }\rho_0(\mathbf{r})

Reading. The ground-state electron density alone, not the full many-electron wavefunction, in principle contains everything needed to determine a system's ground-state energy and properties, and the Kohn–Sham self-consistent scheme provides a practical, computationally tractable route to finding that density.

Scope. The Hohenberg–Kohn theorems are exact statements about the true, unknown \(E_{xc}\); every practical DFT calculation instead uses an approximate functional (Step 4), so DFT results carry a functional-dependent uncertainty that a purist wavefunction-based method, in principle systematically improvable toward the exact answer, does not carry in the same way.

Corollaries & converses

basis-sets, needed to represent the Kohn–Sham orbitals \(\psi_i\) numerically (Step 2), are chosen using essentially the same considerations already developed for Hartree–Fock calculations, since both methods share the identical self-consistent-field, orbital-based computational machinery.
potential-energy-surfaces calculations, which require repeatedly evaluating a system's energy at many different nuclear geometries, are made tractable for larger molecules specifically because DFT's cost scales much more favourably with system size than high-level wavefunction methods, at the price of the functional-dependent accuracy noted in the Result's Scope.
Choosing among LDA, GGA, and hybrid functionals (Step 4) is not a purely formal decision: hybrid functionals like B3LYP are generally preferred for organic thermochemistry, while different functional families may be favoured for solids, transition metals, or weakly bound (dispersion-dominated) systems.

Fails without

Suppose the ground-state density did not uniquely determine the external potential (violating the first Hohenberg–Kohn theorem): more than one distinct physical system could then share an identical density, making "the energy as a functional of density" ill-defined and undermining the entire premise of working with density in place of the wavefunction.
Use a local or semi-local exchange-correlation functional (e.g. plain LDA) for a system dominated by weak, long-range dispersion interactions: such functionals systematically under-bind dispersion-dominated complexes, since they have no built-in mechanism to capture the genuinely non-local correlation responsible for that binding — a well-known, functional-specific failure mode.

Common errors

Describing DFT as an approximate method "in the same way" Hartree–Fock is approximate; the Hohenberg–Kohn theorems (Hypotheses) are formally exact, and all of DFT's approximation is isolated specifically in the unknown exchange-correlation functional (Step 4), a conceptually different source of error from Hartree–Fock's neglect of correlation entirely.
Treating the Kohn–Sham orbitals \(\psi_i\) and orbital energies \(\varepsilon_i\) as though they were directly physically meaningful in the same sense as Hartree–Fock orbitals; they are, strictly, auxiliary quantities of a fictitious non-interacting reference system, useful and often chemically intuitive in practice, but not rigorously guaranteed to carry the same physical interpretation.
Assuming any DFT result is automatically reliable without regard to which exchange-correlation functional was used; different functionals can give meaningfully different answers for the same system, particularly for properties like reaction barrier heights or weak intermolecular interactions.
Forgetting that DFT, unlike a systematically improvable wavefunction hierarchy, has no single, well-defined path to the exact answer; improving a DFT calculation generally means choosing a better functional, not simply adding more computational effort within a fixed method.

Discussion

Pierre Hohenberg and Walter Kohn published the two foundational theorems bearing their names in 1964, and Walter Kohn, together with Lu Jeu Sham, published the practical self-consistent-field scheme (Step 2) the following year in 1965. Walter Kohn was awarded a share of the 1998 Nobel Prize in Chemistry specifically for this development of density functional theory.

A conceptually important, and initially counterintuitive, feature of the theory is that the mapping from density to external potential (Hypotheses) is a purely formal, existence-only guarantee: the Hohenberg–Kohn theorems prove that a functional relationship between \(\rho\) and the ground-state energy exists and is unique, without providing any constructive recipe for writing that functional down exactly — the entire subsequent development of exchange-correlation functionals (Step 4) is an ongoing, largely empirical and semi-empirical effort to approximate a mathematical object known to exist but not known in closed form.

Common misconception: that DFT is simply "cheaper Hartree–Fock." The two methods differ fundamentally in what object they solve for (density versus wavefunction) and in how electron correlation is treated (approximately built into \(E_{xc}\) from the start in DFT, versus entirely absent in bare Hartree–Fock and requiring separate post-Hartree–Fock correlation methods to recover); DFT's favourable cost scaling is a practical consequence of this different formulation, not merely a computational shortcut applied to the same underlying theory.

Worked examples

\text{Geometry optimisation of a small organic molecule using B3LYP with a modest basis set}

A typical DFT workflow: choose a hybrid functional such as B3LYP (Step 4, mixing exact exchange with GGA-type exchange-correlation) and a basis set (basis-sets) to represent the Kohn–Sham orbitals; solve the Kohn–Sham equations self-consistently (Step 3) at a starting geometry; use the resulting energy and its gradient with respect to nuclear positions to step toward lower-energy geometries (potential-energy-surfaces), repeating the full self-consistent-field solution at each new geometry until the gradient vanishes at a stationary point. A

\text{Self-consistent Kohn–Sham density at each geometry} \ \Rightarrow\ \text{energy and gradient} \ \Rightarrow\ \text{optimised structure}

Reading. A full molecular geometry optimisation is simply the Kohn–Sham self-consistent-field procedure of Step 3, repeated at a sequence of nuclear geometries chosen by an optimisation algorithm working from the computed energy gradient at each step.

Scope. The same workflow, with an appropriately chosen functional and basis set, applies to computing reaction energies, transition-state geometries, and vibrational frequencies, essentially the full range of applications potential-energy-surfaces covers.

Problems

State which two quantities in the Kohn–Sham energy expression (Step 1) are treated exactly, and which single term carries essentially all the method's approximation.
Solution
The non-interacting kinetic energy \(T_s[\rho]\), the classical Coulomb repulsion \(J[\rho]\), and the external-potential interaction term are all evaluated exactly (given the density); the exchange-correlation term \(E_{xc}[\rho]\) is the one piece with no known exact closed form and is where every practical approximation (LDA, GGA, hybrid, Step 4) enters.
Explain why the Kohn–Sham equations must be solved self-consistently rather than in a single, direct pass, referencing Step 3.
Solution
The effective potential \(v_{\text{eff}}\) that appears in the Kohn–Sham equations is itself built from the density \(\rho\), which is exactly the quantity those equations are being solved to determine (via the resulting orbitals, Step 2). Solving the equations once with a guessed starting density therefore does not, in general, reproduce that same starting density as output; the calculation must be iterated (rebuild \(v_{\text{eff}}\) from the newest density, resolve, repeat) until the input and output densities agree to within a chosen tolerance, i.e. until self-consistency is reached.
A researcher runs the same molecule's geometry optimisation with two different exchange-correlation functionals and obtains two somewhat different bond lengths. Explain why this outcome is expected, referencing the Hypotheses and the Result's Scope.
Solution
The Hohenberg–Kohn theorems guarantee an exact relationship between the true ground-state density and the exact energy functional, but the true \(E_{xc}\) is unknown; every real calculation substitutes one of many available approximate functionals (Step 4), each built with different underlying approximations and empirical fitting. Since different functionals are, in effect, different approximate theories, some functional-dependent variation in computed properties like bond length is an expected consequence of DFT's Scope, not evidence that either calculation contains an error.