Spin glasses resemble a capricious dish: atomic magnets interact, and finding the perfect combination is tricky. The creators of the quantum algorithm QAOA hoped that quantum superposition would allow them to sift through options faster, as if tasting all spices at once. Drawing on work by Wigner from quantum optics, in computer simulations quantum states were replaced by rotating spin arrows, while the inevitable jitter from the uncertainty principle (Heisenberg) was kept as a random pinch of noise. It turned out that this chaos doesn't speed up the search: too much noise throws the system into disorder, too little prevents trying new things. The optimal level grows with the number of steps, but even then the speed is no higher than with classical methods. Surprisingly, if you completely remove the quantum randomness and re-tune the parameters, the classical version handles it faster, like a precise recipe without shaking the pan. After fixing the result, the conclusion is clear: for this puzzle, quantum supremacy was not confirmed. Perhaps in other tasks a quantum computer will still prove itself, but in the world of entangled magnets, the classical tuning is more accurate for now.
Interestingly, the mathematics of spin glasses migrated from physics to neural network theory: machine learning algorithms also seek equilibrium in a labyrinth of data.