Density functional theory (DFT)

Density functional theory (DFT) is a computational quantum mechanics method used to calculate the electronic structure of atoms, molecules, and solid materials. By approximating how electrons interact with each other and with nuclei, DFT can predict a wide range of properties: formation energies, band gaps, magnetic moments, reaction barriers, bond lengths, and more.

DFT occupies a particular niche in computational science: it’s accurate enough to be scientifically meaningful for many properties, and fast enough (relative to higher-accuracy methods like coupled cluster) to be applied to reasonably large systems. A typical DFT calculation for a small-to-medium molecule or unit cell takes minutes to hours on a modern CPU cluster.

The AI-DFT relationship:

Most AI materials discovery and molecular property prediction models are trained on datasets generated by DFT. The Materials Project, OQMD, and AFLOW databases — which together contain millions of calculated structures and properties — were built by running DFT at scale. AI models (especially graph neural networks like those in GNoME) learn to approximate DFT calculations at a fraction of the compute cost: a prediction that would take hours in DFT takes milliseconds in an AI model.

This creates a hierarchy: DFT provides labeled training data; AI provides fast approximate screening; DFT (or experiment) validates the most promising AI candidates.

Limitations of DFT itself:

• DFT systematically underestimates band gaps for many semiconductors
• Accuracy depends heavily on the exchange-correlation functional chosen — no single functional works well for everything
• Scales poorly with system size; very large systems (>1000 atoms) require approximations or alternative methods
• Doesn’t capture dynamic effects well; reaction rates and finite-temperature behavior require additional frameworks (molecular dynamics, transition state theory)

Related terms: Formation Energy, Active Learning

Related guide: Materials Science