Formation energy

Formation energy (more precisely, the enthalpy of formation) measures how much energy is released or absorbed when a material is synthesized from its component elements in their most stable standard forms. A negative formation energy means the compound is thermodynamically more stable than its constituent elements — a necessary (though not sufficient) condition for a material to exist stably.

In computational materials science, formation energy per atom is typically expressed in electronvolts per atom (eV/atom). Materials with strongly negative formation energies are more likely to be synthesizable and stable under ambient conditions.

Why it matters for AI materials discovery:

Formation energy is the primary property that large-scale AI materials screening models — GNoME, OQMD, the Materials Project — predict and use to filter candidate structures. A material has to be thermodynamically stable before any of its other properties (conductivity, magnetism, catalytic activity) are worth measuring. AI models trained on DFT-computed formation energies can screen millions of hypothetical structures to identify the small fraction worth synthesizing.

How it’s computed:

Traditionally calculated using density functional theory (DFT), which is accurate but computationally expensive — typically hours to days per structure. AI models (graph neural networks trained on DFT datasets like the Materials Project) can predict formation energy in milliseconds at approximately DFT accuracy, enabling screening of orders-of-magnitude more candidates.

Caveats:

• Thermodynamic stability (negative formation energy) is necessary but not sufficient — a material also needs to be kinetically accessible, meaning it must be possible to actually synthesize it under realistic conditions
• DFT formation energies have systematic errors depending on the functional used; models trained on one dataset may not transfer well to another

Related terms: Active Learning, Graph Neural Network

Related guide: Materials Science