Champagne is bottled under ~6 atm of dissolved CO₂ and poured into a glass at 1 atm, leaving it strongly supersaturated. Classical nucleation theory sets a critical radius r* = 2γ/ΔP below which an embryo bubble collapses under surface tension faster than gas can diffuse in. Homogeneous nucleation in the open liquid essentially never clears that barrier — real bubbles nucleate at pre-existing gas pockets trapped in cellulose fibers (from the glass-drying towel) or scratches on the glass, which act as ready-made embryos and only need a radius above r* to fire continuously.
Every defect site in this scene is scored by exp(−(r*/r_site)²): oversized pockets fire almost every frame, undersized ones stay silent — the same selectivity that makes only a handful of a real glass's fibers into active bubble trains. Growing bubbles gain radius by CO₂ diffusion (r ∝ √t) and rise under real Stokes drag, v = (2/9)g(ρ_liquid−ρ_gas)r²/μ, so each bubble visibly speeds up as it grows on its way to the surface.
Equations. Young–Laplace: ΔPbubble = 2γ/r, so a gas pocket survives only if r > r* = 2γ/(Pdissolved − Patm) (γ ≈ 0.05 N/m). Henry's law gives the dissolved pressure from the CO₂ content. Diffusion-limited growth: dr/dt = K/r ⇒ r² = r₀² + 2Kt. Stokes rise: v = 2(ρl−ρg)g r²/(9μ).
Honest simplifications. A site fires with probability exp(−(r*/rsite)²) instead of a full kinetic model; real bubbles carry a contaminated, partly rigid surface and leave the Stokes regime (Re > 1) as they grow; the collective flow induced by many bubbles is not modelled.
What to try. Warm the champagne to 20 °C and watch ΔP and r* change; set wall roughness to 0.2 µm to silence the glass; click the wall to etch a bubble train.