Loading 3D assets...
...
Drag to orbit - wheel / pinch to zoom - try the heat ramp

Why Tiny Gold Melts Early: Nanoparticle Size, Surface Atoms and Melting-Point Depression

A real face-centred-cubic gold cluster, atom by atom. Shrink it and the surface atoms take over: the particle melts hundreds of kelvin below bulk gold. Two textbook models, one picture.
Atoms in the particle-
Surface atoms (coordination < 12)-
Melting point, Gibbs-Thomson-
Melting point, Lindemann-
Bulk gold melts at-
State at this temperature-
Melting point vs radius: Gibbs-Thomson (yellow), Lindemann with counted surface atoms (green), bulk (grey), current temperature (blue band). Dots mark the particle shown.
Independent numerical checks
Each line compares two separate calculations; a FAIL would show here.
The science

Why size matters. A spherical gold particle of radius R has a surface-to-volume ratio of 3/R. A 1 nm particle has about 65 % of its atoms at the surface (counted below on a real face-centred-cubic lattice), a 6 nm particle about 13 %. Surface atoms have fewer neighbours (coordination below 12), so they are bound less tightly and vibrate more.

Model 1: Gibbs-Thomson (thermodynamic). Equating the chemical potentials of a solid sphere and its melt gives Tm(R) = Tbulk (1 − 2σsl / (ρs ΔHf R)), with ρsΔHf = 1.23×109 J/m³ for gold (19.3 g/cm³, 12.55 kJ/mol; typical values). The interface energy σsl is a model parameter: published values scatter a lot, so it is a slider. Real calorimetry on supported gold particles shows melting depression of the same order, but depends on the substrate and shape, so treat the curves as a trend, not a prediction for a specific sample.

Model 2: Lindemann (vibrational). A crystal melts when the rms vibration amplitude reaches a fraction δL of the nearest-neighbour distance. In the high-temperature Debye model ⟨u²⟩ = 9ℏ²T / (m kB θD²), with θD = 165 K for gold. Calibrating δL at the bulk melting point (1337 K) gives δL ≈ 0.114, inside the usual 0.1 to 0.15 range. If surface atoms vibrate β times more (mean-square), the particle average reaches the criterion at Tm = Tbulk / (1 + fs(β − 1)), where fs is the counted surface fraction.

The 3D picture. Every atom of a real FCC cluster is drawn (a = 0.408 nm). Atoms jitter with the Debye amplitude, exaggerated three times so you can see it; surface atoms jitter more by √β. Colour: bright = surface atom, darker = core, orange = liquid-like (its Lindemann ratio is above δL). Surface atoms turn liquid-like first, at Tbulk/β (surface pre-melting); the core follows at the Lindemann melting point. This sequence is a toy rule drawn from the model, not a molecular-dynamics result.

Checks. (1) Atom count against the continuum volume estimate. (2) Surface-atom count by two different algorithms (outermost-neighbour radius vs explicit hash-set neighbour lookup). (3) The Lindemann melting point in closed form against a numerical bisection over the counted atoms. (4) δL inside the usual range. (5) Gibbs-Thomson tends to the bulk value for a 1 micrometre particle.

Not modelled. Real shapes (icosahedra, truncated octahedra), ligands and support, anharmonic vibration, size-dependent latent heat, hysteresis between melting and freezing. No claim is made about any particular experiment. The page is an educational model of two standard ideas.

The heater stage, holder and atom mesh were modelled and baked in Blender; the lattice, statistics and rendering run in Three.js r160.

Made with 🟠 Blender · Three.js