Each domino is a rigid slab of height h, thickness t, width w that rotates about its front bottom edge by an angle θ. Moment of inertia about that edge: I = m (h² + t²) / 3. Gravity gives I θ'' = m g (h/2 sinθ − t/2 cosθ), so a domino only falls forward once tanθ > t/h. Raising its centre of mass from h/2 to √(h²+t²)/2 costs the tipping energy Etip = ½ m g (√(h²+t²) − h), only about 0.3 mJ for a 48 mm domino, while toppling flat releases ½ m g (h − t), about 50 times more. That surplus is what keeps the wave going.
Neighbours meet when the top front corner of domino k reaches the back face of k+1, at roughly sinθc = gap / h (so a gap larger than h can never be bridged). The contact is a frictionless impulse along the face normal with restitution e: J = −(1+e) vn / (a1²/I1 + a2²/I2), where a is the lever arm of the contact about each pivot. Energy lost in impacts and at the table is added up, so total energy can be audited at any time.
Dimensional analysis: with everything scaled by a factor λ, times scale as √λ and speeds as √λ, so v = √(g h) · f(gap/h, t/h, e). A larger domino row is faster, not just longer. The speed-versus-gap curve rises as the gap grows (each domino gets a longer run-up) and then falls again because the tilt needed to reach the next domino costs time; beyond about one height the wave stops.
What is simplified: one rotational degree of freedom per domino (no sliding or twisting), no slipping at the pivot, contact only between neighbours, rigid bodies, a constant restitution e (the real value depends on speed and surface), wood density 700 kg/m³ assumed, and an S-curve is treated as a straight row with the same spacing along the path. Real rows add friction, air, sound and misalignment, so use the absolute speed as a model number, and the trends (scaling, optimum gap, failure at large gaps) as the physics lesson.
3D assets (dominoes with baked pips, wooden table, ruler, steel ball) modelled and baked in Blender.