Modes. For a square plate of side L the standing-wave shapes are approximated by W(u,v) = cos(nπu)cos(mπv) ± cos(mπu)cos(nπv), u, v in [0,1]. The two signs are degenerate here (same frequency); a real plate splits them slightly, and the exciter position decides which mix you get. The difference form always has a nodal line along the diagonal. Note that the old page's mode (3,3) was identically zero (n = m cancels) — it is no longer offered.
Frequency. A thin (Kirchhoff) plate has ρh ∂2w/∂t2 + D∇4w = 0, D = Eh3/12(1−ν2), so ω = √(D/ρh) k2 with k2 = π2(n2+m2)/L2, i.e. f = (π/2)√(D/ρh)(n2+m2)/L2. For a 0.20 m, 1.5 mm aluminium plate (E = 69 GPa, ν = 0.33, ρ = 2700 kg/m3) that is about 91 Hz × (n2+m2). Real free-edge plates follow a different law (Waller/Ritz, roughly f ∝ (n2+m2) only approximately), so treat frequencies as indicative.
Resonance. Amplitude of mode k ∝ 1/√((1−r2)2 + (r/Q)2), r = f/fk, Q = 25 assumed. Away from resonance the plate barely moves: sweep slowly and the sand only organises within about f/Q of a mode.
Sand. Grains hop only where the plate acceleration exceeds g, Γ = G|W| > 1; they scatter with a bias toward lower |W| and stop on the nodal lines. Drift and scatter strengths are illustrative.
Sound. The Sound button plays a pure sine at the drive frequency through Web Audio (check: an offline render is analysed for its spectral peak). Laptop speakers are poor below about 150 Hz; keep the volume low.
Simplifications: plate motion is exaggerated and slowed to 2 Hz (real amplitudes are micrometres); only the antisymmetric/symmetric partner pair is shown, not a mix; no damping variation, air coupling or grain-grain interaction.
Running numerical checks...