The Glass Transition: When a Liquid Forgets How to Flow

The billion-fold slowdown: what actually happens on cooling

Start with a liquid above its melting point T_m and cool it. If nucleation is suppressed — because you cool fast, or the melt resists ordering (network formers like SiO₂, B₂O₃, or good glass-formers like o-terphenyl and polymers) — the liquid slips below T_m without crystallizing. It is now a supercooled liquid, metastable but still ergodic: molecules explore all configurations, just more slowly.

The controlling quantity is the structural (α) relaxation time τ, the time for the liquid to forget its configuration and flow. Through the Maxwell relation it ties directly to viscosity:

• η = G∞ · τ, where G∞ is the instantaneous (high-frequency) shear modulus, typically G∞ ≈ 1–10 GPa for molecular liquids.
• At T_m, τ ≈ 10⁻¹² s and η ≈ 10⁻³ Pa·s (water-like).
• Cooling toward T_g, τ climbs 14 orders of magnitude. When τ ≈ 100 s (η ≈ 10¹² Pa·s, using G∞ ≈ 10¹⁰ Pa), the system can no longer equilibrate within an experiment — it falls out of equilibrium. That fall-out point is T_g.

Because T_g is defined by comparing τ to your patience, it is not a material constant. Cool 10× slower and the liquid stays equilibrated to a slightly lower temperature, so T_g drops by roughly 3–5 K per decade. This rate dependence is the fingerprint that the glass transition is kinetic, not thermodynamic.

The VFT law and why Arrhenius fails

If molecular motion were a simple activated hop over a fixed barrier E, viscosity would obey the Arrhenius law η = η₀·exp(E/k_B T), a straight line on a log-η vs 1/T plot. Some glass-formers — pure silica, germania — do roughly this; they are called strong liquids, with an apparent activation energy near E ≈ 500–700 kJ/mol (several eV per molecule).

Most liquids curve dramatically instead, obeying the empirical Vogel–Fulcher–Tammann (VFT) law (Vogel 1921, Fulcher 1925, Tammann–Hesse 1926):

• η(T) = η₀ · exp[ B / (T − T₀) ]
• η₀ is a high-temperature prefactor (~10⁻⁴ Pa·s), B a constant with units of kelvin, and T₀ the Vogel temperature where η formally diverges.
• Critically T₀ < T_g — typically T₀ ≈ 0.7–0.8 T_g — so the divergence lurks below the transition but is never reached, because the system freezes first.

The equivalent for relaxation time is τ = τ₀·exp[B/(T − T₀)]. The divergence at finite T₀ means the barrier to flow effectively grows without bound: the effective activation energy E_eff(T) = k_B B·T²/(T − T₀)² blows up as T → T₀. Something collective, not a single fixed barrier, governs the arrest.

Fragility: strong versus fragile liquids

Austen Angell's insight (1985) was to plot log₁₀η against T_g/T — the scaled Angell plot — so every liquid passes through the same point (log₁₀η = 12) at T_g/T = 1. The steepness at that point defines the fragility index:

• m = d(log₁₀τ) / d(T_g/T) evaluated at T = T_g.
• Strong liquids (SiO₂, GeO₂) have m ≈ 20 — nearly Arrhenius, gently curving lines. Their tetrahedral network resists reorganization uniformly.
• Fragile liquids (o-terphenyl m ≈ 81, many polymers m ≈ 100–200) collapse steeply: viscosity plummets over a narrow temperature range near T_g.

Fragility reflects how quickly the liquid's landscape of accessible configurations shrinks on cooling. Strong liquids have a rigid, energetically simple potential-energy landscape; fragile liquids have a rugged one whose deep basins trap the system abruptly. Fragility correlates with the jump in heat capacity ΔC_p at T_g — fragile liquids show a large step (o-terphenyl ~ 110 J/mol·K), strong liquids barely a ripple.

Free volume, cooperative regions, and the physical picture

Why should flow require diverging cooperation? Two complementary pictures dominate:

• Free-volume theory (Cohen–Turnbull, 1959): a molecule can only move when a neighboring void larger than some critical v* opens up. The probability scales as exp(−γv*/v_f), and the free volume v_f shrinks linearly toward zero at T₀ — recovering the VFT form. Below T_g the free volume is frozen at a nonequilibrium value.
• Cooperatively Rearranging Regions (Adam–Gibbs, 1965): as T falls, molecules can only move in ever-larger correlated clusters of size z*. Adam–Gibbs give τ = τ₀·exp[C/(T·S_c)], where S_c is the configurational entropy. When S_c → 0, τ diverges — again reproducing VFT if S_c ∝ (T − T₀).

Experiments confirm a growing dynamic correlation length: near T_g cooperative clusters span only ξ ≈ 5–20 molecular diameters (~1–3 nm), and a handful of molecules must move together for any one to move at all. This is why the transition is broad and static structure looks unchanged: X-rays see the same amorphous halo, but the dynamics have become spatially heterogeneous — some regions are 10⁵× more mobile than others just nanometers away.

The Kauzmann paradox and a hidden ideal glass

Walter Kauzmann (1948) noticed something unsettling. Because a supercooled liquid loses entropy faster than its crystal (its C_p is larger), extrapolating the liquid's excess entropy downward, it would fall below the crystal's — reaching zero excess entropy at the Kauzmann temperature T_K, typically only 10–20 K below T_g. A liquid with less entropy than a crystal at the same temperature violates the Third Law's spirit.

• Remarkably, T_K ≈ T₀ for most fragile liquids — the entropy catastrophe and the viscosity divergence point to the same temperature.
• This suggests a hypothetical ideal glass transition at T_K: a genuine thermodynamic transition to a unique, entropy-crisis-avoiding amorphous ground state, forever hidden because kinetic arrest at T_g always intervenes first.

Whether that ideal transition truly exists, or whether the extrapolation simply breaks down, remains debated. Random First-Order Transition theory and Replica theory predict it; dynamic-facilitation theories deny it. No experiment can reach T_K because equilibrating there would take longer than the age of the universe — the ultimate uncooperative liquid.

Aging, the fictive temperature, and everyday glass

Below T_g a glass is out of equilibrium and slowly ages: it keeps relaxing toward the (unreachable) equilibrium liquid line, densifying by parts per thousand over years. Its state is bookmarked by a single number, the fictive temperature T_f — the temperature at which the frozen structure would be in equilibrium. Rapidly quenched glass has a high T_f (open, low density); annealed glass has a low T_f. This is why optical-fiber preforms and precision lenses are annealed for hours: to lower T_f, tighten the network, and stabilize the refractive index (Δn ~ 10⁻⁴ matters).

• Window glass (soda-lime) has T_g ≈ 550 °C; the myth that medieval panes are 'thicker at the bottom because glass flows' is false — at room temperature τ exceeds 10³² years. The thickness is a manufacturing artifact of the crown-glass process.
• Metallic glasses (e.g. Zr–Cu–Al) need cooling rates of 10⁶ K/s (or bulk formers ~1–100 K/s) to beat crystallization; they are twice as strong as steel because there are no dislocations or grain boundaries.
• Polymers, amber, hard candy, and cryopreserved cells are all glasses. Pharmaceutical amorphous drugs are vitrified deliberately — the disordered state dissolves faster than the crystal.

From gorilla-glass phone screens to the vitrified sugars that let tardigrades survive desiccation, the glass transition is the physics of trapping a liquid's disorder in a solid's rigidity — a billion-fold slowdown frozen in place.

Crystallization versus vitrification: two fates of a cooling liquid

Property	Crystallization (freezing)	Glass transition (vitrification)
Type	1st-order thermodynamic transition	Kinetic arrest, cooling-rate dependent
Volume/enthalpy	Discontinuous drop at T_m	Continuous; slope (thermal expansion, C_p) drops
Latent heat	Yes, released at T_m	None released
Structure	Long-range periodic order (Bragg peaks)	Amorphous; frozen liquid short-range order
Signature	Sharp T_m, independent of rate	Broad T_g, shifts ~3–5 K per decade of rate
Viscosity at transition	Low (~1–10 Pa·s)	≈ 10¹² Pa·s