Write descend(momentum, steps=30, lr=0.09) on the narrow valley f(w1, w2) = w1² + 10·w2² from (5, 1): gradients (2·w1, 20·w2), velocity update v = m·v + g per axis, step along velocities. For momentum 0.0, 0.5, 0.9 print f"momentum={m} dist={dist:.4f}" (distance from the optimum), then print whether 0.5 beats BOTH neighbors — the sweet spot, not a free lunch.