Two players take a bad beat. One is only a little annoyed. One is fully, completely tilted. Classic intuition says the calmer one settles down first. Sometimes it's wrong β the more tilted player's calm-probability curve can shoot past the mildly-annoyed one on the way back to baseline. That's a real Mpemba effect, running on real multi-state relaxation math, not a scripted gag.
A plain rung-by-rung ladder (ΞΊ=0) is a totally positive chain: whoever starts more tilted stays stochastically more tilted forever β no crossing is possible, ever, for any rates. The shortcut edges are what let the two players' relaxation "modes" mix with different signs, which is exactly the mechanism the paper identifies in the kinesin motor's 6-state cycle.
This is the paper's real quantity of interest: not "how calm right now" but how far each player's whole mood distribution still is from the table's long-run baseline. Each curve is a sum of several decaying exponential modes β when the coefficients on those modes have opposite signs for A vs B, the sum stops being monotone in a simple way and the two distances can swap order.
The chain's generator Q is diagonalized (5 relaxation eigenmodes, one of them the
steady state with rate 0). Each player's initial distribution is projected onto every mode, giving
an overlap coefficient c_k. The full trajectory is
p(t) = Ξ£_k c_k Β· e^(Ξ»_k t) Β· Ο_k. Watch the signs of c_k below flip between
A and B on the faster modes β that sign flip, combined with those modes still being non-negligible at
early times, is exactly what produces the crossover (or its absence).
| Mode | Relaxation rate |Ξ»| | c_k (Player A) | c_k (Player B) |
|---|