# Nucleation **Nucleation** is the first step of a first-order [[Phase_transition|phase transition]]: the appearance, somewhere in the parent phase, of a cluster of the new phase large enough to keep growing. It is needed because a small cluster is mostly surface. Forming it releases a bulk free energy proportional to its volume but costs an interfacial energy proportional to its area, and below a certain size the area term wins, so the cluster dissolves again.[^porter-ch4][^callister-ch10] Nucleation is therefore an activated process, and the delay it imposes is why water can be [[Supercooling|supercooled]], why a [[Solid_solution|solid solution]] can be [[Quenching|quenched]] and held, and why almost every [[Microstructure|microstructure]] a metallurgist cares about is a frozen accident of how many nuclei appeared and when. In the microsim below the single control is the undercooling `ΔT` below the melting point. The sim draws the classical curve `ΔG(r) = 4·π·r²·γ − (4/3)·π·r³·ΔG_v`, with the volumetric driving force `ΔG_v = ΔH_f·ΔT/T_m`, and reads off its maximum: the critical radius `r* = 2·γ/ΔG_v` and the barrier `ΔG* = 16·π·γ³/(3·ΔG_v²)`. The copper preset uses the row's values `γ = 0.177 J/m²` and `ΔH_f = 1.63×10⁹ J/m³` with `T_m = 1358 K`.[^cu-presets] As `ΔT` grows, both `r*` and `ΔG*` collapse — `r*` as `1/ΔT` and `ΔG*` as `1/ΔT²` — and the nucleation rate, which follows `I ∝ exp(−ΔG*/(k·T))`, does not creep upward but switches on.[^turnbull-fisher1949] On the [[Materials_science]] flagship this page serves Part VI, *Fundamentals › Kinetics*, immediately after [[Fick's_laws_of_diffusion|Fick's laws]] and immediately before [[Spinodal_decomposition|spinodal decomposition]], the transformation that needs no nucleus at all. ## Characteristics Everything classical follows from the two competing terms. Differentiating `ΔG(r)` and setting the result to zero gives `r* = 2·γ/ΔG_v`; substituting back gives `ΔG* = 16·π·γ³/(3·ΔG_v²)`.[^porter-ch4] A cluster smaller than `r*` lowers its energy by shrinking and a cluster larger than `r*` lowers its energy by growing, so `r*` is an unstable equilibrium — a ridge, not a valley — and the critical nucleus is the one configuration in the whole problem that is trying to fall apart in both directions at once. The numbers explain why nucleation looks like a threshold. Take the sim's copper preset at `ΔT = 50 K`: `ΔG_v = 1.63×10⁹·(50/1358) = 6.0×10⁷ J/m³`, so `r* = 5.9 nm` and `ΔG* = 2.6×10⁻¹⁷ J`, about 161 eV. At 1308 K that is 1,429 times `k·T`, and `exp(−1429)` is not a small number but a nonexistent one. Push the undercooling to 236 K and `ΔG_v` rises to `2.8×10⁸ J/m³`, `r*` falls to 1.25 nm, `ΔG*` to `1.16×10⁻¹⁸ J` (7.2 eV) and the exponent to about 75 — small enough that with the usual attempt frequency the rate becomes measurable. The critical nucleus shrinks with it: a 5.9 nm copper sphere holds roughly 73,000 atoms, a 1.25 nm sphere about 700. Turnbull's droplet measurements found that pure metals generally supercool to roughly `0.18·T_m` before freezing, about 240 K for copper, which is very close to where this arithmetic says the rate turns on.[^turnbull1950] Classical nucleation theory assumes the cluster is a sphere with the bulk phase inside and the flat-interface energy `γ` outside, neither of which is true for an object 700 atoms across, and measured rates can differ from the prediction by many orders of magnitude even when the trend is right.[^porter-ch4] What survives every criticism is the shape of the answer: a barrier falling as `1/ΔG_v²` inside an exponential. ### Heterogeneous nucleation often dominates homogeneous nucleation Homogeneous nucleation, in which the cluster forms in the bulk with no help, almost never happens in practice. If the new phase can instead form as a cap on an existing [[Surface_science|surface]] — a mould wall, an oxide particle, a [[Grain_boundary|grain boundary]], a speck of dust — part of its own interface is paid for by destroying interface that was already there, and the barrier falls by a purely geometric factor `S(θ) = (2 + cos θ)·(1 − cos θ)²/4`, where `θ` is the contact angle the new phase makes with the substrate.[^porter-ch4][^callister-ch10] `S(θ)` is brutal. At `θ = 90°` it is 0.5, halving the [[Activation_energy|barrier]]; at 60° it is 0.156; at 30° it is 0.0129, a factor of 78. Applied to the copper case above, an undercooling of 50 K that gives a hopeless homogeneous exponent of 1,429 gives `1429 × 0.0129 = 18` on a substrate the melt wets to 30°, which nucleates freely. The critical radius is unchanged — the cap has the same curvature as the free sphere — but the cap contains only 1.3 % of the atoms, about 940 instead of 73,000. This is why a melt poured into a mould freezes within a few degrees of its melting point while a droplet of the same metal suspended in an inert medium will supercool hundreds. It is also an engineering lever: [[Casting|casting]] practice adds inoculants that present a well-wetted crystallographic surface, so that many nuclei form early and the [[Crystallite|grains]] of the [[Alloy|alloy]] stay small.[^porter-ch4] The same argument, read backwards, explains boiling chips, cloud condensation nuclei and the cellulose fibres on which the bubbles in a glass of sparkling wine form. ### Computer simulation studies of simple models Simulation is the only way to watch a critical nucleus directly, and it is hard for exactly the reason nucleation is interesting: a barrier of 60 `k·T` means the event is astronomically rare on the timescale a molecular-dynamics run can reach. Standard practice is therefore to bias the sampling — umbrella sampling along a cluster-size coordinate, or forward-flux sampling — and reconstruct the unbiased rate afterwards. The models used are deliberately crude: hard spheres, Lennard-Jones fluids and the [[Ising_model|Ising model]] under Metropolis dynamics, in which a field applied against the current magnetisation makes the ordered domain the "new phase" and its droplets nucleate exactly as classical theory describes.[^anag-ch15] The Ising testbed also supplies the lattice machinery the Wikitube sim set reuses: single-spin flips accepted with probability `min(1, exp(−β·ΔE))`, on a grid large enough that a critical droplet is many sites across.[^anag-ch15] The recurring result of such studies is that the classical picture gets the exponential right and the prefactor wrong, and that real pathways are often two-step — a dense disordered cluster forms first and orders afterwards — which a one-coordinate theory cannot represent. ### Spinodal region Metastability has a limit. Inside a two-phase region the parent phase is metastable: its free energy is above the two-phase tie line, but it sits in a local minimum and only a nucleus can get it out. Push far enough and the curvature of the free-energy curve changes sign, and beyond that line — the spinodal — there is no local minimum and no barrier at all, so the phase separates by the continuous growth of infinitesimal fluctuations rather than by nucleation.[^likharev-sm4] The [[Van_der_Waals_equation|van der Waals]] loop shows both regions on one isotherm: between the coexistence volumes and the turning points the fluid is metastable, and Likharev's account notes that metastable branches survive only to near the spinodals, water being supercoolable to about −50 °C and superheatable to about +270 °C before the region runs out.[^likharev-sm4] Crossing the spinodal changes the kinetics completely, which is the subject of [[Spinodal_decomposition|spinodal decomposition]]; the two mechanisms produce different microstructures — isolated particles of fixed composition on one side, an interconnected modulated structure of growing amplitude on the other. ## Nucleation of crystals When the new phase is crystalline, the driving force can be written from whatever is displacing the system from equilibrium. Cooling a melt below `T_m` gives `ΔG_v = ΔH_f·ΔT/T_m`, the sim's expression, which is simply the latent heat multiplied by the fractional undercooling.[^porter-ch4] Cooling or concentrating a solution gives `ΔG_v = (k·T/v)·ln S`, with `S` the [[Supersaturation|supersaturation]] ratio and `v` the molecular volume, and a vapour gives the same form with the pressure ratio. In every case `γ` sits in the barrier as a cube while the driving force sits there as a square, so interfacial energy is by far the more powerful variable — halving `γ` cuts `ΔG*` by a factor of eight. ### Primary and secondary nucleation Primary nucleation is the appearance of a crystal where none of that phase exists: homogeneous in the bulk, or heterogeneous on a foreign surface. Secondary nucleation is the production of new crystals by crystals already present — fragments knocked off by stirrer impacts and crystal–crystal collisions, or ordered clusters shed from the boundary layer around a growing face. In industrial crystallisers secondary nucleation usually dominates, because it needs only a modest supersaturation, and that is what makes seeding effective: a seeded batch controls its own crystal count instead of leaving it to whatever dust the vessel happened to contain. ### Experimental observations on the nucleation times for the crystallisation of small volumes Measuring homogeneous nucleation requires removing the heterogeneous sites, and the standard trick is subdivision. Turnbull dispersed a metal into many droplets a few micrometres across, so that the few effective nucleants present were isolated in a small minority of droplets and the rest had to nucleate on their own; the bulk of the population then supercooled to a reproducible temperature near `0.18·T_m`.[^turnbull1950] Such experiments also expose the statistics. Nucleation is a Poisson process: the chance that a droplet of volume `V` is still liquid after time `t` is `exp(−I·V·t)`, so identical droplets held at identical undercooling crystallise at wildly different times, and the observable is a survival curve rather than a single temperature.[^turnbull-fisher1949] Because `I` depends exponentially on `ΔG*`, which depends on `1/ΔT²`, the survival curve is extremely sharp in [[Temperature|temperature]] and very broad in time — a few degrees separate "never" from "immediately", but at a fixed undercooling the waiting times spread over decades. ### Ice Water is the most-studied case and the least typical. Pure water in small droplets can be cooled far below 0 °C before it freezes — the Portal Book's account of metastability puts the practical supercooling limit near −50 °C — because the ice–water interfacial energy is high relative to the latent heat and the barrier stays large until deep undercooling.[^likharev-sm4] In bulk water, and in the atmosphere, freezing is instead controlled by whatever heterogeneous nucleants are present: mineral dust, soot, and biological ice nucleators, which raise the freezing temperature by presenting a surface that [[Ice|ice]] wets well. The consequences run from cloud physics to food. Rapid freezing produces many nuclei and fine ice crystals; slow freezing produces few nuclei and large ones, which is why the same ice cream is smooth or gritty depending on the freezer. In the microsim the ice case is a preset change and nothing more: substitute water's `γ`, `ΔH_f` and `T_m` and the same curve answers the question. ## Examples Nucleation appears wherever a phase has to be born rather than merely grown, and the three classic settings differ mainly in which term of `ΔG(r)` is largest. ### Nucleation of fluids (gases and liquids) A bubble must push back the liquid against its own [[Surface_tension|surface tension]], and the Laplace pressure across a bubble of radius `r` is `ΔP = 2·γ/r`. For water, with `γ ≈ 0.072 N/m`, a 1 µm bubble needs 1.4 bar of excess pressure inside it and a 10 nm bubble needs 144 bar, which is why truly homogeneous [[Boiling_point|boiling]] requires superheats of well over a hundred degrees and never occurs in a kettle. Real boiling nucleates in micrometre-sized cavities in the heating surface that trap gas and so start with a large `r`; smooth, degassed surfaces produce the violent bumping familiar from a laboratory flask. [[Cavitation|Cavitation]] is the same process driven by tension instead of heat, and [[Evaporation|condensation]] in a vapour is its mirror image, with the [[Vapor_pressure|vapour pressure]] over a small droplet raised by curvature according to the Kelvin equation.[^kelvin] ### Nucleation of crystals In [[Casting|casting]] the nucleation count sets the grain size and therefore much of the mechanical behaviour: many nuclei give fine [[Equiaxed_crystal|equiaxed]] grains and a stronger, more isotropic product, few nuclei give coarse columnar ones. Grain refiners are added for that reason alone. In solution crystallisation the same lever controls particle size distribution, and in [[Crystal_growth|crystal growth]] for [[Semiconductor|semiconductors]] the goal is the opposite extreme — exactly one nucleus, the seed, and no others anywhere in the melt. ### Nucleation in solids Solid-state nucleation adds a term the liquid case does not have. A new crystal inside a crystal is generally the wrong size and shape for the hole it occupies, so an elastic strain energy `ΔG_s` per unit volume opposes the transformation, and the driving force becomes `ΔG_v − ΔG_s`.[^porter-ch5] Nuclei therefore form preferentially on defects that both relieve strain and destroy existing interface: [[Grain_boundary|grain boundaries]], [[Dislocation|dislocations]], and the coherent boundaries of earlier precipitates. Coherent nuclei with a low `γ` are favoured early even when they are not the equilibrium phase, which is why [[Precipitation_hardening|age hardening]] in aluminium–copper alloys proceeds through [[Guinier–Preston_zone|Guinier–Preston zones]] before any equilibrium precipitate appears.[^porter-ch5] Because solid-state nucleation needs [[Diffusion|diffusion]] and diffusion slows exponentially as the temperature falls, the transformation rate is a product of a term rising with undercooling and a term falling with it. The result is the C-shaped curve of a time–temperature–transformation diagram, fastest at an intermediate temperature and slow at both ends, and the overall transformed fraction follows the [[Avrami_equation|Avrami]] form `X = 1 − exp(−k·t^n)`.[^avrami][^porter-ch5] Quenching past the nose of the C faster than nucleation can occur is exactly how [[Martensite|martensite]] is obtained, and martensite itself is the exception that proves the rule: it is diffusionless, nucleates on pre-existing defects almost athermally, and grows at the speed of sound.[^porter-ch6] *See also:* [[Classical_nucleation_theory]] — the full rate expression behind the sim's `exp(−ΔG*/(k·T))` · [[Supersaturation]] · [[Crystal_growth]] · [[Ostwald_ripening]] · [[Supercooling]] · [[Avrami_equation]] · [[Casting]] · [[Spinodal_decomposition]] · [[Phase_transition]] · [[Miscibility_gap]] ## References [^gibbs]: Gibbs, J. W. (1876–1878). "On the Equilibrium of Heterogeneous Substances." *Transactions of the Connecticut Academy of Arts and Sciences* 3 (page to pin). The surface/volume competition that the whole subject rests on is Gibbs's. [^volmer-weber1926]: Volmer, M.; Weber, A. (1926). "Keimbildung in übersättigten Gebilden." *Zeitschrift für physikalische Chemie* 119: 277–301. [^becker-doring1935]: Becker, R.; Döring, W. (1935). "Kinetische Behandlung der Keimbildung in übersättigten Dämpfen." *Annalen der Physik* 416 (8): 719–752. [^turnbull-fisher1949]: Turnbull, D.; Fisher, J. C. (1949). "Rate of Nucleation in Condensed Systems." *The Journal of Chemical Physics* 17 (1): 71–73 (the `I = I_0·exp(−ΔG*/(k·T))` form and the attempt-frequency prefactor). [^turnbull1950]: Turnbull, D. (1950). "Formation of Crystal Nuclei in Liquid Metals." *Journal of Applied Physics* 21 (10): 1022–1028. https://doi.org/10.1063/1.1699435 The droplet-dispersion method and the finding that pure metals supercool by roughly 0.18 of their absolute melting temperature; `0.18 × 1358 K ≈ 240 K` for copper is computed here from that fraction, not read from the paper. [^porter-ch4]: Porter, D. A.; Easterling, K. E.; Sherif, M. Y. *Phase Transformations in Metals and Alloys*, 3rd ed. (2009), Ch. 4 Solidification: homogeneous nucleation in melts, `ΔG_v = ΔH_f·ΔT/T_m`, `r* = 2γ/ΔG_v`, `ΔG* = 16πγ³/(3ΔG_v²)`, heterogeneous nucleation and the shape factor `S(θ)`, inoculation and grain refinement (page to pin). [^porter-ch5]: Porter, Easterling & Sherif (2009), Ch. 5 Diffusional Transformations in Solids: the strain-energy term, nucleation on grain boundaries and dislocations, coherent interfaces and Guinier–Preston zones, TTT diagrams and the Avrami equation (page to pin). [^porter-ch6]: Porter, Easterling & Sherif (2009), Ch. 6 Diffusionless Transformations: martensite nucleation on pre-existing defects and its athermal character (page to pin). [^callister-ch10]: Callister, W. D.; Rethwisch, D. G. *Materials Science and Engineering: An Introduction*, 9th ed. (2014), Ch. 10 Phase Transformations: the free-energy-versus-radius construction, the critical radius and barrier, heterogeneous nucleation, growth, and the Avrami relation (page to pin). [^likharev-sm4]: Likharev, K. *Essential Graduate Physics, Part SM: Statistical Mechanics* (2013), Ch. 4 Phase Transitions, pp. 107–142, and in particular the van der Waals loop, the mechanically unstable segment and the spinodals at pp. 108–113: the model cannot say whether the condensed phase is liquid or solid, and metastable branches survive only to near the spinodals, with water quoted as supercoolable to about −50 °C and superheatable to about +270 °C (page to pin for the individual statements). https://open.umn.edu/opentextbooks/textbooks/part-sm-statistical-mechanics [^anag-ch15]: Anagnostopoulos, K. *Computational Physics: A Practical Introduction to Computational Physics and Scientific Computing (using C++)*, 2nd ed. (2016), Ch. 15 Simulation of the d = 2 Ising Model, pp. 520–593 (Metropolis acceptance `min(1, exp(−β·ΔE))`, the lattice update loop and domain growth). https://open.umn.edu/opentextbooks/textbooks/computational-physics-a-practical-introduction-to-computational-physics-and-scientific-computing-using-c [^avrami]: Avrami, M. "Kinetics of Phase Change. I–III." *The Journal of Chemical Physics* 7 (1939), 8 (1940) and 9 (1941) (page to pin). [^kelvin]: Thomson, W. (1871). "On the equilibrium of vapour at a curved surface of liquid." *Philosophical Magazine*, series 4, volume 42 (page to pin). [^cu-presets]: The copper presets used by the M32 sim — `γ = 0.177 J/m²`, `ΔH_f = 1.63×10⁹ J/m³`, with `T_m = 1358 K` — are the row's values. No Portal Book is a materials text, and the tabulated source for the solid–liquid interfacial energy and volumetric latent heat of copper (Porter's Table 4.1 or an equivalent handbook entry) still needs pinning. Every number derived from them in this article — `ΔG_v`, `r*`, `ΔG*`, the exponents and the atom counts — was computed here from those presets and copper's lattice parameter, and is not a book value. <!-- MATTERSIM:BEGIN g24 — Matter & Energy Cluster microsim (framework build, specs/sims/Nucleation.json); do not hand-edit inside --> **Microsim — three.js (Wikitube framework), pending deploy:** *Nucleation* will play here once `https://wikitube-3d-microsims.netlify.app/matter/Nucleation.html` is live. <!-- pending: <div class="wt-sim" data-src="https://wikitube-3d-microsims.netlify.app/matter/Nucleation.html" data-title="Nucleation"></div> --> <!-- MATTERSIM:END --> ## Wikipedia : Wikitube **Strict pair:** [Wikipedia](https://en.wikipedia.org/wiki/Nucleation) : [Wikitube](https://en.wikitube.io/wiki/Nucleation) · pinned revision [1371069234](https://en.wikipedia.org/w/index.php?oldid=1371069234) · 2026-09-11 ## Previous hub tags Hubs: `Life_Physics`. Portals: [[PORTAL_Materials_science]]. --- *Matter & Energy Cluster child articles, wave 1 · 2026-09-11 · drafted · Materials_science row M32 · sim pending (matter/Nucleation).*