Parabolic Grain Growth Law

A grain boundary has energy, roughly half a joule per square metre for a high-angle boundary in a metal. A polycrystal therefore has a thermodynamic incentive to reduce its total boundary area, and at a temperature where atoms can move it does exactly that: boundaries migrate, large grains eat small ones, and the mean grain size grows. Beck and co-workers fitted the kinetics in a 1948 Transactions of the AIME paper, and the result is a power law — Dⁿ minus the starting size to the same power is proportional to the time.

The n = 2 case is the ideal one, and it can be derived. The driving pressure on a curved boundary is proportional to its curvature, which for a grain of diameter D goes as 1/D. If boundary velocity is proportional to driving pressure through a mobility, then dD/dt ∝ 1/D, and integrating gives D² − D₀² = Kt. Real alloys almost never measure 2. They measure 3 or 4, and the reason is always that something is dragging on the boundary: solute atoms segregated to it that must be dragged along, or second-phase particles that pin it.

That pinning is the Zener effect, and it is one of the most exploited phenomena in steelmaking. A boundary passing a particle must create new boundary area to get past it, so each particle exerts a retarding pressure. When the retarding pressure from a dispersion of particles equals the driving pressure from curvature, growth stops entirely at a limiting grain size of roughly the particle radius divided by the volume fraction. No power law can represent a growth process that stops, which is why fitting the parabolic law across a hold long enough to reach the limit gives a nonsense exponent. Microalloyed steels use exactly this: tenths of a percent of niobium, titanium or vanadium form fine carbonitrides that pin austenite boundaries during reheating and hot rolling, holding the austenite fine so that the ferrite it transforms to is fine as well.

K is unit-bound and its dimensions change with n. This site takes K in µmⁿ per second — grain size in micrometres, time in seconds. At n = 2 that is µm²/s; at n = 3 it is µm³/s, a numerically unrelated quantity. A K quoted against a different exponent, a different length unit, or a time in hours cannot be used here, and the error it produces looks entirely plausible. K also belongs to one temperature only: it follows an Arrhenius law K = K₀exp(−Q/RT), so a constant measured at 1000 °C says nothing usable about 950 °C without the activation energy alongside it.

The engineering consequence follows straight from Hall–Petch, and it is the reason to care. Coarsening the grain lowers the yield strength as the inverse square root, and it lowers the toughness at the same time — the ductile-to-brittle transition temperature of a ferritic steel rises sharply with grain size, which is a much more dangerous outcome than the loss of strength. That combination is what makes an over-long or over-hot soak so damaging, and what limits the heat input in welding: the coarse-grained region of a heat-affected zone, right beside the fusion line, has spent time at a temperature where nothing pins the boundaries, and it is routinely the toughest place in a weld to satisfy an impact requirement.

One last practical note on measuring D at all. Mean grain size is a statistical quantity and it depends on the method: the ASTM E112 grain size number, the mean linear intercept, and the equivalent-circle diameter from an area count are three different numbers for the same microstructure, related by geometric factors. A growth constant fitted with one measure and applied with another is off before it starts. State the method with the number.